Abstract
Recommender systems — the algorithmic architectures that determine what content users encounter on digital platforms — have been analyzed extensively as instruments of misinformation, political polarization, and attention capture. This paper proposes an additional and underexamined frame of analysis: recommender systems as mood enclosure, the progressive confinement of users within affective monocultures whose boundaries are determined not by the user’s considered values or formative intentions but by the engagement-optimization imperatives of commercial platform design. Drawing on affective science, ecological theory, the history of land enclosure as an analogy for the privatization of common experiential resources, and the conceptual frameworks developed across the preceding papers in this series, the paper argues that the affective monoculture produced by engagement-optimized recommendation is a structural harm distinct from — and in some respects more foundational than — the informational and political harms more commonly analyzed. It then proposes three classes of countermeasure — diversity quotas in content recommendation, narrative sequencing modes as an alternative curation paradigm, and user-set formation goals as a mechanism for value-aligned platform engagement — and develops each in sufficient theoretical and practical detail to support policy, design, and advocacy application.
I. Introduction
A. The Enclosure Analogy
Between the fifteenth and nineteenth centuries, the common lands of England — the open fields, meadows, and wastes that had supported the subsistence agriculture and common life of rural communities for centuries — were systematically enclosed: legally privatized, physically fenced, and converted from shared communal resources to individually owned productive property. The enclosure movement generated significant agricultural productivity gains, contributed materially to the industrial revolution, and produced, as its social consequence, the dispossession of millions of rural people from the common resources upon which their way of life depended. What had been held in common — and whose common holding had sustained a particular texture of communal, economic, and ecological life — was converted into private property optimized for a single purpose: productive yield.
The analogy to algorithmic content curation is neither perfect nor merely rhetorical. What recommender systems have enclosed is the common affective range of human experience — the broad, varied, culturally shared field of emotional registers, narrative forms, and experiential textures that prior media environments, for all their limitations, made available to individuals whose attention moved across them without algorithmic mediation. The pre-algorithmic media environment was not neutral or benign; it had its own distortions, exclusions, and commercial pressures. But it did not possess the technical capacity to identify each individual user’s precise affective preferences and to continuously refine a content environment calibrated to keep the user maximally engaged within a progressively narrowing experiential territory. Algorithmic recommendation has this capacity, has deployed it at global scale, and has produced, as its affective consequence, something that deserves the name enclosure: the conversion of a shared, varied, common experiential field into privately managed monocultures optimized for yield — yield measured in engagement, attention-minutes, and behavioral data.
The enclosure analogy also illuminates the character of the harm in ways that purely informational or psychological framings do not. Enclosure is not merely an individual harm; it is a commons harm. The affective monoculture produced by algorithmic enclosure does not only impoverish the individual user whose emotional range is progressively narrowed. It impoverishes the shared affective commons — the collective capacity for mutual emotional understanding, for grief and celebration held in common, for the shared navigation of difficulty and ambiguity — that a community requires to function as more than an aggregation of isolated, privately managed experiential units. The harm is ecological as well as individual, and its remedies must be designed at both scales.
B. Scope and Relationship to Prior Work
This paper is the fifth in a series developing a theoretical and applied account of friction, formation, and the conditions of genuine human flourishing in digital environments. Prior papers established the psychological mechanisms by which mood-optimization behavior erodes emotional resilience (Paper 1), the systems-ecological function of friction as a structural buffer (Paper 2), the conceptual distinction between affective teleportation and formative transition zone traversal (Paper 3), and a set of humane friction design patterns capable of building rather than merely blocking capacity (Paper 4). This paper addresses the infrastructural level at which all of these dynamics operate — the algorithmic recommendation layer that determines which experiential content users encounter and in what sequence — and proposes interventions at that level.
The formation dependency thesis established in Paper 3 is foundational here: genuine formation of any kind — moral, relational, spiritual, psychological — requires the traversal of transition zones, and the traversal of transition zones requires encounter with experiential content sufficiently varied and sequenced to include the full range of registers that formation demands. Algorithmic enclosure of the affective field is therefore not merely a wellbeing problem; it is a formation problem. A person whose content environment has been narrowed to the emotional registers that maximize their engagement is a person who has been deprived of the experiential raw material that moral, relational, and spiritual formation requires. The stakes are not trivial.
II. Recommender Systems and the Production of Affective Monoculture
A. The Architecture of Enclosure
Modern recommender systems are built on a common architectural foundation: collaborative filtering, content-based filtering, or hybrid approaches that combine both, all optimized against behavioral engagement signals — clicks, watch time, shares, return visits, and the micro-behavioral data generated by cursor position, scroll velocity, and interaction latency (Ricci et al., 2015). The optimization target is not user satisfaction, not user wellbeing, not the alignment of content experience with user values, and not the development of user capacities over time. It is engagement — the behavioral trace of attention capture — because engagement is what the business model monetizes.
The affective consequence of this optimization target is structural rather than incidental. Engagement-optimized systems learn, with high precision and at scale, which emotional registers drive which users’ engagement behavior most reliably. High-arousal, high-valence negative states — outrage, anxiety, contempt, fear — consistently drive higher engagement metrics than low-arousal positive states — contentment, quiet gratitude, reflective appreciation — because they activate the behavioral urgency systems that produce the clicking, sharing, and return-visiting that engagement metrics capture (Berger & Milkman, 2012). Algorithmic systems optimized for engagement are therefore, by structural necessity, systems that progressively amplify high-arousal negative affect and progressively marginalize the lower-arousal, more complex, more ambivalent registers that constitute much of the richest human emotional experience.
This is the mechanism of affective enclosure: not a deliberate design decision to harm users’ emotional range, but an emergent consequence of the optimization target that the business model requires, operating at scale and over time, producing the progressive narrowing that individual session analysis would not detect but longitudinal affective pattern analysis consistently reveals. Coyne (2016) documented the narrowing of emotional register in algorithmically curated media consumption across a multi-year longitudinal study; the narrowing was not experienced by users as loss — they reported satisfaction with their content — but was measurable in the progressive reduction of affective variety across their consumption patterns. The enclosure is comfortable, which is precisely what makes it effective and what makes its harm so difficult to perceive from within it.
B. Monoculture as Ecological Category
The term monoculture is borrowed from agricultural ecology, where it describes the practice of cultivating a single crop species across a large area of land. Monoculture produces impressive yields under benign conditions and catastrophic vulnerability under stress: the Irish potato famine, caused by a blight that swept through the genetically homogeneous potato monoculture that British agricultural policy had established as the primary food source of the Irish peasantry, is the paradigmatic historical case. The monoculture’s efficiency under benign conditions and its catastrophic fragility under novel stress are two expressions of the same structural feature: the elimination of the diversity that buffering, resilience, and adaptation require.
Applied to the affective domain, monoculture produces the same pattern: comfortable consistency under benign conditions and catastrophic fragility when life demands engagement with emotional registers that the monoculture has eliminated. The person whose affective environment has been algorithmically narrowed to high-arousal engagement registers has not merely lost aesthetic variety; they have lost the practiced familiarity with grief, ambivalence, quiet contemplation, and patient endurance that the hard passages of human life require. They encounter those passages with an affective immune system that has never been exercised in the relevant register, and the result is precisely the reduced tolerance, shortened transition zones, and lower emotional elasticity documented in Paper 1 of this series.
The ecological frame also illuminates the commons dimension of the harm. Agricultural monoculture does not merely impoverish the individual field; it degrades the broader agricultural ecosystem — the pollinator populations, the soil microbiome, the watershed dynamics — that diverse cultivation supports. Affective monoculture at scale does not merely impoverish individual emotional range; it degrades the shared affective commons that communities require for their collective emotional life: the shared capacity for communal grief, for collective moral deliberation that can inhabit uncomfortable ambiguity, for the civic patience that democratic life demands, and for the mutual intelligibility that genuine community requires. When millions of individuals have had their affective range narrowed by the same algorithmic logic, the community they constitute has lost something it did not know it held in common and cannot easily recover by individual action alone.
C. Feedback Dynamics and Lock-In
The affective monoculture is self-reinforcing in ways that distinguish it from other forms of content homogenization. Standard filter bubble analysis focuses on informational homogeneity — the tendency of recommender systems to show users content that confirms their existing beliefs and excludes content that challenges them (Pariser, 2011). The affective dimension adds a layer of lock-in that informational analysis misses: the narrowing of affective range produces a reduced tolerance for affective contrast that makes algorithmically diverse content not merely unpreferred but experientially intolerable.
The dynamic works as follows. Early exposure to engagement-optimized content trains the user’s content preference signals toward high-arousal registers. Subsequent recommendations, shaped by these signals, deliver more high-arousal content. Extended engagement with high-arousal content reduces the user’s tolerance for lower-arousal registers — not through deliberate preference but through the use-it-or-lose-it mechanism of affective tolerance documented in Paper 1: the capacity that is not exercised atrophies, and the atrophied capacity makes the content that would exercise it feel uncomfortable, slow, or unrewarding. The user’s revealed preference data now strongly signals preference for high-arousal content, not because this reflects their considered values or their genuine affective range, but because the algorithm has, through the progressive narrowing dynamic, produced a user whose practiced tolerance genuinely is limited to the registers it has amplified.
This feedback dynamic is critical to understand because it means that revealed preference data — the primary signal that recommender systems optimize against — is not a neutral measure of user values or genuine interests. It is a measure of the affective range that the system has already produced in the user. Optimizing against it deepens the enclosure further. The user who reports, genuinely and accurately, that they prefer high-arousal content and find quieter, more contemplative content unrewarding has not expressed a free preference; they have reported the affective range that the system has narrowed them to. The distinction between genuine preference and algorithmically produced preference is not merely philosophical; it is the central design and ethical challenge of any countermeasure proposal.
D. Differential Vulnerability
The affective monoculture is not produced uniformly across all users, and its differential distribution follows patterns of particular ethical concern. Research on engagement-optimized content consistently finds that users who are younger, who have higher baseline anxiety or depression levels, who use platforms more intensively, and who have less developed pre-existing self-regulatory capacity are more vulnerable to the narrowing dynamic and more severely affected by its outcomes (Twenge, 2017). This differential vulnerability means that algorithmic enclosure functions, in effect, as a regressive system: it takes most severely from those who can least afford the loss.
The vulnerability of younger users deserves particular emphasis. Adolescence is the primary developmental window for the formation of emotional range, affective identity, and the self-regulatory capacities that allow adult navigation of the full spectrum of human experience. An algorithmic environment that narrows the affective range of adolescent users during this window does not merely reduce their current wellbeing; it shapes the affective architecture they will carry into adulthood, with consequences for their relational capacity, moral formation, and resilience that extend far beyond the platform’s terms of service. Haidt and Rausch (2022) documented the correlation between smartphone and social media adoption patterns across age cohorts and measurable increases in adolescent anxiety, depression, and reduced emotional resilience — a correlation that is most plausibly explained, at least in part, by the affective monoculture mechanism described here.
III. Three Classes of Countermeasure
A. Diversity Quotas in Content Recommendation
1. Conceptual Basis
The diversity quota is the most structurally direct countermeasure to affective monoculture: a design constraint that requires recommender systems to include, within each unit of content delivery — each session feed, each recommendation batch, each playlist — a specified minimum proportion of content in emotional registers other than those the engagement-optimization signal would select. It is, in the language of agricultural ecology, a crop rotation requirement: a structural mandate for the affective variety that maximized yield optimization systematically eliminates.
The diversity quota is not a new concept in recommender system design. Informational diversity requirements have been proposed and, in some regulatory contexts, implemented as countermeasures to filter bubble dynamics (Helberger et al., 2018). What is proposed here extends the diversity requirement from the informational to the affective domain — from the diversity of political perspectives, sources, and factual claims to the diversity of emotional registers, narrative modes, and affective textures that a formation-supporting content environment requires.
The conceptual basis for this extension draws on the ecological resilience literature reviewed above, the formation dependency thesis of Paper 3, and the cross-register suggestion pattern developed in Paper 4. The diversity quota is, in effect, the structural implementation of the cross-register suggestion at the platform architecture level rather than the individual interaction level: rather than offering individual users optional cross-register suggestions, it mandates affective diversity as a property of the recommendation architecture itself. This architectural implementation addresses the limitation of individual-level interventions — that they depend on user engagement and choice — by building diversity into the system’s default output.
2. Design Specifications
An affective diversity quota system requires, as its technical foundation, an affective taxonomy of content — a classification system that maps content across an emotional register space with sufficient granularity to distinguish the registers that engagement-optimization conflates. This is not a trivial technical requirement, but it is not novel: affective content analysis is a well-developed field, and the application of affective classification to content recommendation has been proposed and prototyped in academic research contexts (Tkalčič & Chen, 2015).
A workable affective taxonomy for diversity quota purposes might include the following primary register dimensions: arousal level (high to low), valence (positive to negative), complexity (simple to ambivalent or mixed), temporal orientation (present-focused to historically grounded to future-contemplative), and social orientation (individual to communal). Content classified across these dimensions can be mapped in an affective space, and diversity quotas can be specified as constraints on the distribution of recommended content across this space — requiring, for example, that no single arousal-valence quadrant constitute more than a specified percentage of a session’s content, or that each recommendation batch include at least one piece of content in each of a specified set of primary registers.
The quota specifications themselves require calibration for context, user age, platform type, and the specific formation goals — if any — that the user has set through the formation goal mechanism described in Section C below. A platform serving primary-age children requires different quota specifications than one serving adult professionals; a platform dedicated to music listening requires different specifications than one dedicated to news and commentary. The principle is constant — affective diversity is structurally mandated rather than left to engagement-optimization — while the specific parameters are context-dependent.
The relationship between the diversity quota and the user’s current affective state requires careful design attention. A quota that rigidly delivers content in registers the user is not currently able to access — requiring a user in acute distress to encounter challenging contemplative content, for example — is not a capacity-building tool but an affective imposition that violates the graduated exposure principle established in Paper 4. Effective quota implementation therefore includes both static minimum diversity requirements (which apply regardless of current state) and dynamic modulation mechanisms (which adjust the register distance of required diversity content based on detected current affective state, providing the adjacency that genuine cross-register traversal requires).
3. Policy and Regulatory Dimensions
The diversity quota countermeasure has implications at the policy and regulatory level that extend beyond individual platform design decisions. Regulatory frameworks for platform content — currently focused almost exclusively on harmful content removal and informational diversity requirements — could be extended to include affective diversity requirements as a condition of platform operation, analogous to the nutritional labeling and composition requirements that food regulation applies to the dietary environment. Just as food regulation does not merely prohibit toxic ingredients but requires the disclosure of nutritional composition and, in some jurisdictions, mandates minimum nutritional standards, platform regulation could require affective composition disclosure and mandate minimum affective diversity standards.
This is not a straightforward regulatory proposal, and its development would require extensive engagement with the technical, political, and rights-based complexities of content regulation that are beyond this paper’s scope. What the paper asserts is the principled basis for such regulation: if recommender systems systematically produce affective monocultures whose harms are demonstrable at both individual and societal levels, the regulatory frameworks that govern the media environment have a legitimate interest in those systems’ affective outputs analogous to their legitimate interest in the informational and safety outputs that existing frameworks address.
B. Narrative Sequencing Modes
1. Conceptual Basis
The narrative sequencing mode is an alternative content curation paradigm that replaces the engagement-optimization algorithm’s logic — surface what will most reliably keep this user engaged in this moment — with a different organizing principle: sequence content as a deliberate narrative arc, with a beginning, development, and resolution structured to move the user through a meaningful experiential journey rather than to maintain them in a state of maximum engagement intensity.
The conceptual foundations of narrative sequencing are developed in detail in Papers 3 and 4 of this series, particularly in the discussions of experiential arc completion and musical sequencing as a formation technology. The present section extends those foundations to the platform architecture level, proposing narrative sequencing as a system-wide curation paradigm rather than a feature available at the individual interaction level.
Narrative sequencing draws on a tradition of curatorial practice — in music programming, exhibition design, liturgical sequencing, and educational design — that understands the ordering of experience as itself a formative act. The museum curator who sequences an exhibition to move visitors from familiar territory through challenge to expanded understanding is performing a formative act. The liturgical musician who sequences a service from lament through confession to assurance and celebration is performing a formative act. The teacher who designs a course to move students from confidence through productive confusion to reconstructed, deeper understanding is performing a formative act. In each case, the sequencing is not incidental to the experience’s value; it is, in large part, the source of it. The engagement-optimization algorithm is constitutionally incapable of this kind of sequencing because its optimization target — maximum momentary engagement — is incompatible with the deliberate introduction of the discomfort, contrast, and resolution-deferral that meaningful narrative arcs require.
2. Design Specifications
A narrative sequencing mode would be available as an explicit alternative to standard engagement-optimized recommendation, selectable by the user as a primary or secondary curation paradigm. Its implementation requires several components that differ fundamentally from engagement-optimization architecture.
The first component is a narrative grammar — a formal description of the arc structures that the sequencing mode is designed to produce. Different narrative grammars are appropriate for different platform contexts and user formation goals. A music platform might implement narrative grammars derived from classical concert programming principles: the contrast and balance of tempo, mode, and register across a listening session; the movement from accessible entry to challenge to reward that characterizes the best concert programs. A news and commentary platform might implement narrative grammars derived from good journalistic practice: the movement from immediate news through contextual background to analytical depth to reflective commentary that allows a user to move from reaction to understanding across a session. An educational platform might implement narrative grammars derived from instructional design: the sequence from prior knowledge activation through challenge through struggle through resolution and consolidation that the productive failure literature identifies as the structure of genuine learning.
The second component is a session arc manager — an algorithmic function that maps a user’s session in real time against the narrative grammar being applied, identifies the current position in the arc, and selects subsequent content to advance the arc toward its next phase rather than to maximize immediate engagement. The session arc manager is, in design terms, the algorithmic embodiment of the curator — the function that holds the overall shape of the experience in view while making individual content selection decisions, rather than optimizing each decision independently against an engagement signal.
The third component is arc transparency — a user interface element that makes the session’s narrative arc visible to the user: where they are in the arc, what phase is coming next, and how much of the arc remains. This transparency serves multiple functions: it makes the narrative sequencing mode’s operation legible, supports user formation goal alignment, provides a natural framework for arc completion behavior, and offers the kind of temporal orientation that reduces the anxious, undirected quality of passive engagement-optimized consumption. A user who can see that they are in the challenge phase of a deliberately sequenced session and that resolution is coming is in a fundamentally different experiential and formative relationship to their content than a user who is simply inside an undifferentiated feed of maximally engaging material.
The fourth component is arc completion support — the implementation of the complete-the-arc option developed in Paper 4 at the session level rather than the individual content level. The narrative sequencing mode creates natural arc completion points — moments at which a narrative arc has reached its resolution phase and the user is offered a genuine session conclusion rather than an immediate redirect to new content. These completion points are the designed analog of the natural pauses that traditional media formats — the end of a film, the final track of an album, the last paragraph of an essay — provided before continuous autoplay eliminated them.
3. Narrative Sequencing and Liturgical Wisdom
The narrative sequencing mode has a structural analogy in the sequencing logic of traditional liturgical practice that deserves explicit acknowledgment, given the formation-oriented framework of this series. The liturgical sequence — understood across traditions as a deliberate ordering of communal experience through phases of approach, acknowledgment, confession, assurance, instruction, response, and sending — is among the most sophisticated narrative sequencing systems in human cultural history, refined across millennia of practice in the service of a clearly understood formation goal: the reorientation of the whole person toward right relationship with God and neighbor.
The wisdom embedded in liturgical sequencing is not merely traditional; it reflects a sophisticated practical understanding of how human beings are formed through structured communal experience over time — the same understanding that the formation dependency thesis formalizes in psychological and systems-theoretical terms. The movement from approach to lament in the Hebrew psalmic tradition, from confession to assurance in Reformed liturgical practice, from darkness to light in the paschal sequence, are not arbitrary conventions. They are, in formation terms, precisely calibrated narrative grammars designed to move participants through the transition zones that genuine formation requires — to prevent the teleportation that bypasses the uncomfortable middle registers and delivers only the preferred emotional conclusions.
The relevance to platform design is not that platforms should adopt liturgical forms, but that the wisdom encoded in those forms is available as a resource for narrative sequencing design. The principles that make a liturgical sequence formative — deliberate movement through affective contrast, support for the uncomfortable middle registers, the refusal of premature resolution, the genuine conclusion that supports disengagement and integration — are platform-agnostic principles of formative experience sequencing that can inform the design of narrative sequencing modes without requiring the adoption of their specifically theological content.
C. User-Set Formation Goals
1. Conceptual Basis
The user-set formation goal is the most person-centered of the three proposed countermeasures: a mechanism through which users can specify, in explicit and actionable terms, the formative outcomes they want their platform engagement to support, and through which the platform’s recommendation architecture is aligned with those goals rather than with engagement-optimization imperatives.
The formation goal mechanism rests on a distinction between revealed preference and considered preference that is foundational to both ethical platform design and to classical accounts of practical reason. Revealed preference — what users actually click, watch, and return to — reflects their momentary behavioral responses to algorithmically curated stimuli under conditions of affective narrowing and engagement-optimization pressure. Considered preference — what users would choose if they could specify their engagement patterns against their genuine values and long-term formation goals — is frequently different, sometimes radically so. Research on the preference-behavior gap in media consumption consistently finds that users’ stated preferences for content that challenges, informs, and enriches are significantly more diverse than their behavioral consumption patterns, and that they express retrospective dissatisfaction with passive engagement-optimized consumption at rates that reveal a substantial gap between what they do and what they would prefer to do (Sunstein, 2007).
The formation goal mechanism is designed to close this gap — to provide a structural mechanism by which considered preference can be expressed and operationalized in platform behavior, redressing the current condition in which only revealed behavioral preference shapes algorithmic output. It treats users as agents capable of specifying their own formation goals and deserving of a platform architecture that takes those specifications seriously, rather than as behavioral systems to be optimized for engagement yield.
2. Design Specifications
A formation goal system is implemented as an explicit interface layer — distinct from the standard settings and preferences architecture — through which users can specify their formation intentions across several dimensions.
The first dimension is affective range development: the user specifies which emotional registers they want to develop greater familiarity and tolerance for. The interface presents an accessible description of available registers — without clinical or technical language — and invites the user to identify both registers they want to explore more deeply and registers they recognize they currently avoid. The formation goal system then uses this specification to shape diversity quota requirements and cross-register suggestion content toward the user’s identified developmental edges.
The second dimension is formation domain: the user specifies which formation domains their platform engagement is intended to support. Available domains might include: relational formation (developing empathy, perspective-taking capacity, and understanding of human diversity); intellectual formation (developing critical thinking, tolerance for complexity, and comfort with well-grounded uncertainty); moral formation (developing ethical sensitivity, exposure to moral challenge, and the habit of reflection on values and their application); spiritual formation (developing contemplative capacity, awareness of transcendent meaning, and engagement with questions of ultimate significance); and creative formation (developing aesthetic range, exposure to artistic challenge, and the habit of engaged attention to craft). The formation goal specification in each domain shapes both the content types and the narrative sequencing modes that the recommendation architecture deploys.
The third dimension is temporal formation rhythm: the user specifies the temporal pattern of their formation-oriented engagement — the proportion of each session, each week, or each month they intend to dedicate to formation-oriented versus recreation-oriented content engagement. This specification gives the recommendation architecture a temporal framework for balancing formation-supportive friction with the genuine restorative value of unchallenging content that the user freely chooses. Formation goals are not a mandate for continuous difficulty; they are a framework for ensuring that difficulty and challenge appear in the user’s content environment in the proportions that the user has specified as appropriate to their formation intentions.
The fourth dimension is formation accountability: the user specifies how they want the platform to support their formation goals over time. Options might include: periodic formation reviews — session summaries that relate the user’s actual content engagement to their stated formation goals; progress indicators — accessible, non-gamified displays of how the user’s engagement patterns have aligned with their formation goals over time; and formation companions — the option to share formation goals with a trusted person outside the platform who can provide the relational accountability that platform-level mechanisms cannot replicate.
3. Formation Goals and Autonomy
The formation goal mechanism raises a question that must be addressed directly: does a platform that actively shapes user experience toward specified formation outcomes cross the line from support into paternalism, and if not, where is that line? The answer depends entirely on the sovereignty architecture of the formation goal system, and the sovereignty architecture is therefore the mechanism’s most critical design feature.
A formation goal system that is genuinely voluntary — freely initiated, fully configurable, and easily disabled — respects user autonomy in a way that engagement-optimization does not. Engagement-optimization shapes user experience toward platform-serving outcomes that the user has not chosen and, typically, does not know are operating. Formation goal alignment shapes user experience toward outcomes that the user has explicitly specified, in a system whose operation is fully transparent and fully reversible. The ethical asymmetry between the two is not subtle: one is the covert alignment of the user’s experience with commercial imperatives; the other is the transparent alignment of the platform’s behavior with the user’s own stated values.
The deeper issue is that the concept of neutrality in platform design is unavailable. Every recommendation architecture shapes user experience toward some outcome; the choice is not between shaping and not shaping but between shaping toward engagement yield and shaping toward something the user might actually endorse. The formation goal mechanism makes this choice explicit and gives users genuine agency over it for the first time. That is not paternalism. It is, in the most direct sense available in a commercial platform context, genuine respect for user autonomy.
The formation goal mechanism also has an explicitly relational and communal dimension that individual autonomy framing alone does not capture. Formation goals can be set not only for individual users but for family units — parents specifying formation goals for their household’s shared platform engagement — and for communities — educational institutions, faith communities, and civic organizations specifying formation goal frameworks for their members’ platform engagement within an opted-in community context. This communal formation goal architecture acknowledges that formation is not only an individual project but a communal one, and that the affective commons whose enclosure this paper has diagnosed requires communal as well as individual countermeasures.
IV. Implementation, Resistance, and the Question of Scale
A. The Business Model Problem
None of the three proposed countermeasures is technically complex by current platform engineering standards. Affective classification of content, narrative arc management algorithms, and user preference architecture are all within the current capabilities of major platform engineering organizations. What they require is not technical innovation but business model realignment: the willingness to accept the reduction in short-term engagement metrics that affective diversity, narrative sequencing, and formation goal alignment will produce in exchange for the user wellbeing and long-term trust benefits that the same measures generate.
This willingness is not currently widespread, and it is unlikely to emerge spontaneously from competitive market dynamics that reward engagement metrics and penalize their reduction. The business model problem is therefore not merely a platform design problem but a regulatory and market structure problem: the conditions under which platform design can prioritize user formation over engagement yield require either regulatory mandates that level the competitive playing field — removing the competitive disadvantage of individual platforms that voluntarily reduce engagement — or alternative revenue models that are positively correlated with user wellbeing rather than inversely correlated with it.
The subscription model offers the most promising structural alternative. A platform that monetizes through subscription rather than advertising has, in principle, stronger incentives to produce genuine user value — since subscribers whose wellbeing is served by the platform will maintain their subscriptions, while subscribers whose wellbeing is harmed will not — than a platform that monetizes through advertising, which is positively correlated with engagement metrics regardless of wellbeing effects. The alignment is imperfect — subscription platforms can and do exploit engagement-optimization dynamics — but the structural incentive relationship is meaningfully different, and the regulatory agenda of requiring platforms above a certain scale to offer formation-supportive subscription tiers is a tractable policy proposal that does not require the wholesale transformation of platform business models.
B. The Measurement Problem
A significant obstacle to the adoption of formation-oriented platform design is the measurement problem: the outcomes that formation-oriented design aims to produce — expanded emotional range, developed self-regulatory capacity, deepened moral sensitivity, stronger relational competence — are difficult to measure with the precision and immediacy that engagement metrics provide. Engagement metrics are available in real time at the individual interaction level; formation outcomes develop over months and years and require longitudinal measurement instruments that current platform evaluation practice does not routinely deploy.
Closing this measurement gap requires investment in longitudinal wellbeing research that tracks formation outcomes across extended platform engagement periods — research analogous to the longitudinal health studies that established causal relationships between dietary patterns and health outcomes over decades of follow-up. Several such research programs are currently underway in academic contexts, but their scale and the independence of their methods from platform commercial interests require significant strengthening if their results are to provide the evidentiary basis for regulatory and design action that the scale of the problem warrants.
In the interim, the wellbeing metrics proposed in Paper 4 of this series — retrospective session endorsement, emotional state trajectory across sessions, self-regulatory capacity development, and user preference for formation-supportive features when given genuine choice — provide accessible proxies that can be deployed in current platform evaluation practice. They do not replace longitudinal formation measurement; they provide an available starting point while that measurement infrastructure is developed.
C. The Cultural Precondition
Behind the business model problem and the measurement problem lies a cultural precondition whose importance this series of papers has returned to repeatedly: the widespread cultural internalization of the assumption that discomfort is a malfunction and that the value of any experience lies entirely in its destination state. Diversity quotas, narrative sequencing modes, and formation goals are all, in different ways, proposals for the reintroduction of experiential variety, challenge, and deliberate sequencing into content environments that users have been trained, over years of engagement-optimized consumption, to expect will deliver their preferred affective states on demand. Some users will embrace these proposals; many, at least initially, will not.
This cultural precondition cannot be addressed by platform design alone. It requires the parallel development of the cultural literacy — the understanding of how algorithmic enclosure works, what it costs, and what genuine formation requires — that allows users to make informed decisions about their content environments rather than simply accepting the defaults that engagement-optimization produces. This literacy is a task for education, journalism, community formation, and the kind of sustained public reflection that this series of papers is attempting, in its limited way, to contribute to.
The formation goal mechanism is, among the three proposed countermeasures, the one most directly engaged with this cultural precondition: it treats users as capable of genuine formative self-direction and provides a structural mechanism for the exercise of that capacity. Its effectiveness depends on users who understand what formation is, why it matters, and what the algorithmic environment has been doing to prevent it — users who bring to the formation goal interface a considered understanding of their own formation needs rather than a set of preferences shaped by the enclosure the interface is designed to address.
This is a demanding precondition, and acknowledging its depth honestly is part of the intellectual honesty this series has tried to maintain throughout. The countermeasures proposed here are necessary but not sufficient. They are design and policy contributions to a problem that is, at its deepest level, a formation problem — and the recovery from the affective monoculture that algorithmic enclosure has produced will require, alongside good design and wise policy, the renewal of the communal, spiritual, and relational formation environments that remain, as they have always been, the primary ecology in which genuine human flourishing grows.
V. Conclusion
The concept of algorithmic enclosure names something that has been widely felt and insufficiently theorized: the progressive confinement of human affective experience within privately managed monocultures optimized for yield rather than flourishing. It is not a metaphor for a minor aesthetic impoverishment. It is a description of a structural condition with demonstrable consequences for individual emotional resilience, relational capacity, moral formation, and the collective affective infrastructure that communities require to function as more than aggregations of privately managed attention units.
The three countermeasures proposed here — diversity quotas, narrative sequencing modes, and user-set formation goals — operate at different levels of the problem: the first at the architectural level of what the algorithm is structurally required to deliver, the second at the curatorial level of how content is organized and sequenced, and the third at the agentive level of what the user has specified their engagement to serve. Each is insufficient alone; together, they constitute a coherent approach to the problem’s multiple dimensions, and their combined implementation at scale would represent a significant step toward the recovery of the affective commons that algorithmic enclosure has progressively diminished.
The enclosure of the English commons took centuries to accomplish and produced harms that took generations to become visible and longer still to be adequately named. The enclosure of the affective commons has taken less than two decades and has produced harms that are already measurable, already visible in the longitudinal data, and already felt in the texture of personal, relational, and civic life by anyone who attends carefully to what the quality of human emotional engagement in the algorithmically saturated environment has actually become.
The commons can be recovered. It requires the willingness to name what has been taken, to understand the mechanisms by which it was taken, and to design — with the seriousness, the ethical clarity, and the formation-oriented ambition that the scale of the loss requires — the systems that might, over time, give it back.
Notes
Note 1 — On the Enclosure Analogy and Its Limits: The enclosure analogy is deployed throughout this paper as an illuminating frame rather than a precise historical parallel. Significant disanalogies exist between the enclosure of agricultural commons and the enclosure of the affective commons that should be acknowledged. Physical commons were finite and rivalrous — one person’s use diminished another’s; the affective commons of human emotional range is neither finite nor rivalrous in the same sense. Additionally, the historical enclosure movement involved explicit state action and legal mechanism; algorithmic enclosure is a diffuse, emergent consequence of billions of individual platform interactions shaped by invisible architectural choices. These disanalogies do not undermine the analogy’s central utility — the description of the conversion of a shared resource into a privately managed monoculture optimized for yield — but they should prevent the analogy from being pushed into territory where the differences matter more than the similarities.
Note 2 — On Affective Classification and Technical Feasibility: The diversity quota mechanism proposed in Section III.A requires an affective taxonomy of content that classifies material across emotional register dimensions with sufficient reliability to support quota compliance monitoring. This is presented as technically feasible on the basis of existing affective computing and sentiment analysis research, but the current state of affective content classification should be characterized honestly: classification reliability varies significantly across content types, cultural contexts, and register dimensions, and the misclassification rates of current systems would introduce meaningful noise into any quota compliance framework. The proposal is technically directional rather than immediately deployable; significant research and development investment in affective classification accuracy, bias correction, and cross-cultural calibration would be required before a regulatory affective diversity quota system could be implemented with confidence in its outputs.
Note 3 — On Revealed vs. Considered Preference: The distinction between revealed preference and considered preference drawn in Section III.C is foundational to the formation goal mechanism but raises questions in behavioral economics that this paper does not fully engage. The behavioral economics literature has extensively documented the instability of stated preferences and the many ways in which considered preferences are themselves shaped by framing, availability, and the same cognitive biases that affect revealed preference (Thaler & Sunstein, 2008). The claim here is not that considered preference is perfectly reliable but that it is meaningfully different from revealed preference in engagement-optimized environments, and that a mechanism which gives considered preference structural influence over platform behavior is ethically and practically superior to one that gives only revealed preference such influence. The imperfection of the considered preference measure does not undermine the formation goal mechanism; it requires that the mechanism be designed with appropriate humility about the stability and reliability of the formation goals it operationalizes.
Note 4 — On the Liturgical Analogy in Section III.B: The discussion of liturgical narrative sequencing wisdom as a resource for platform narrative sequencing design is offered as a substantive design resource, not as an ornamental cultural reference. The claim is that the formation wisdom encoded in mature liturgical traditions — the understanding of how human beings are moved through affective and moral transformation by deliberate experiential sequencing over time — is both genuine and platform-relevant, and that the secular design profession’s general unfamiliarity with this literature represents a significant gap in its available resources. Readers interested in pursuing the liturgical sequencing tradition as a design resource might begin with Old Testament scholarship on the Psalter’s arrangement and editorial theology, proceed through the patristic sources on catechetical formation, and engage contemporary liturgical theology literature on the formative function of ordered worship.
Note 5 — On the Relationship Between Individual and Communal Countermeasures: This paper has focused primarily on platform design countermeasures to algorithmic enclosure. It should be explicitly acknowledged that these countermeasures, however well designed and widely implemented, are insufficient responses to a problem whose roots are as much communal and cultural as they are technological. The recovery of the affective commons requires, alongside good platform design, the renewal of the non-digital formation environments — family, friendship, community, worship, craft, outdoor experience, and embodied communal life — that are the primary ecology of genuine emotional and moral formation. Platform countermeasures that support and extend these environments are more valuable than those that attempt to replicate their functions within the platform; the latter risk producing a more sophisticated form of the same substitution problem that engagement-optimized design has created. The best outcome of the formation goal mechanism, in particular, is one in which users specify formation goals that drive them toward more genuine engagement with the non-digital formation environments in their lives — not goals that make the platform itself into a more sophisticated formation institution.
References
Berger, J., & Milkman, K. L. (2012). What makes online content viral? Journal of Marketing Research, 49(2), 192–205. https://doi.org/10.1509/jmr.10.0353
Coyne, S. M. (2016). Technology and the development of emotional and social competence. In L. Balter & C. S. Tamis-LeMonda (Eds.), Child psychology: A handbook of contemporary issues (3rd ed., pp. 378–396). Psychology Press.
Haidt, J., & Rausch, Z. (2022). Social media and mental health: A collaborative review. Unpublished manuscript, New York University Stern School of Business.
Helberger, N., Karppinen, K., & D’Acunto, L. (2018). Exposure diversity as a design principle for recommender systems. Information, Communication & Society, 21(2), 191–207. https://doi.org/10.1080/1369118X.2016.1271900
Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.
Pariser, E. (2011). The filter bubble: What the internet is hiding from you. Penguin Press.
Pennebaker, J. W. (1997). Writing about emotional experiences as a therapeutic process. Psychological Science, 8(3), 162–166. https://doi.org/10.1111/j.1467-9280.1997.tb00403.x
Ricci, F., Rokach, L., & Shapira, B. (Eds.). (2015). Recommender systems handbook (2nd ed.). Springer. https://doi.org/10.1007/978-1-4899-7637-6
Sunstein, C. R. (2007). Republic.com 2.0. Princeton University Press.
Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press.
Tkalčič, M., & Chen, L. (2015). Personality and recommender systems. In F. Ricci, L. Rokach, & B. Shapira (Eds.), Recommender systems handbook (2nd ed., pp. 715–739). Springer. https://doi.org/10.1007/978-1-4899-7637-6_21
Twenge, J. M. (2017). iGen: Why today’s super-connected kids are growing up less rebellious, more tolerant, less happy — and completely unprepared for adulthood. Atria Books.
