# The Skeuomorphic Interval <iframe width="100%" height="20" scrolling="no" frameborder="no" allow="autoplay; encrypted-media" src="https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2394861303&color=%23ff5500&inverse=false&auto_play=false&show_user=true"></iframe><div style="font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;"><a href="https://soundcloud.com/bryantmcgill" title="Bryant McGill" target="_blank" style="color: #cccccc; text-decoration: none;">Bryant McGill</a> · <a href="https://soundcloud.com/bryantmcgill/the-skeuomorphic-interval" title="The Skeuomorphic Interval: Personalization, Provisioning, and Enclosure" target="_blank" style="color: #cccccc; text-decoration: none;">The Skeuomorphic Interval: Personalization, Provisioning, and Enclosure</a></div> ### Personalization, Provisioning, and the Enclosure That Arrives Without Announcement > [!map] Personalization, provisioning, and enclosure > This essay is the canonical synthesis for the [[wiki/Skeuomorphic Interval|skeuomorphic interval]]: the passage from [[wiki/Recommendation Systems|personalization]] through [[wiki/Operational Representation|operational representation]], [[wiki/Affordance Engineering|affordance construction]], [[wiki/Continuous Reconciliation|recursive feedback]], [[wiki/Provisioned Environment|provisioning]], and [[wiki/Enclosure|enclosure]]. [[articles/Cyber-Physical Lessons from Imperial Beekeeping|Cyber-Physical Lessons from Imperial Beekeeping]] supplies the prior model-space architecture; [[articles/AI Escape Is the Wrong Metaphor|AI Escape Is the Wrong Metaphor]] supplies the constructed computational habitat; and [[collections/Machine Succession|Machine Succession]] carries the larger civilizational transition. The argument of this essay is a single progression: **[[wiki/Recommendation Systems|personalization]] becomes [[wiki/Operational Representation|representation]], representation becomes [[wiki/Affordance Engineering|affordance construction]], affordance construction becomes recursive feedback, recursive feedback becomes [[wiki/Provisioned Environment|provisioning]], and provisioning at sufficient breadth becomes [[wiki/Enclosure|enclosure]].** Skeuomorphism is not the thesis. It is the transitional mechanism by which that migration remains socially and cognitively traversable — the reason a civilization can pass through an ontological change while continuing to reason, legislate, invest, and organize in the vocabulary of the arrangement that has already ended. The critical object is therefore not the app, the store, the feed, the job, the twin, or the neural interface considered separately. It is the **progressively individualized actionable environment** assembled around an entity whose responses continuously modify the representation from which the next environment is generated. The mechanism that binds every stage below is one [[wiki/Reconciliation Loop|loop]]. Prediction selects an [[wiki/Affordances|affordance]]; the affordance elicits behavior; the behavior updates the representation; the updated representation determines the next affordance. Each turn of that loop narrows the difference between estimating what someone wants and producing the conditions under which they want it. Nothing in the loop requires a central coordinator, a governing directive, or anyone's intention to construct an enclosure. It requires only that the same persistent representation sit beneath enough separate services, and that each service be permitted to act on it. What the enclosure eventually becomes is not determined by the architecture. It is determined by **selection pressure**: what gets requested, rewarded, financed, retained, measured, legislated, and technologically amplified across billions of local optimizations. A civilization that persistently rewards effortless entertainment, anticipatory gratification, attention capture, and the avoidance of cognitive difficulty trains its institutions and its machines toward one attractor. A civilization that rewards autonomy, reversibility, provenance, epistemic resistance, plural agency, and open-ended discovery selects another. There need be no prime directive for an emergent [[wiki/Objective Function|objective landscape]] to form, and that landscape is the thing actually being decided while everyone argues about interfaces. Claims below are graded rather than leveled. An argument of this reach survives hostile reading only if it declines to convert plausibility into fact, and the strong claims are strong precisely because the weak ones are not smuggled in beside them. ## I. The declarative substrate *Established, deployed, and unglamorous.* The first stage is already behind us, which is why it is invisible. Across infrastructure engineering the unit of specification shifted from procedure to **[[wiki/Desired State|declared desired state]]**, with a [[wiki/Reconciliation Loop|reconciliation loop]] responsible for closing the gap between description and reality. The formal vocabulary arrived in 2022, when the IRTF's Network Management Research Group published its clarification of **intent-based networking**, distinguishing intent from policy and from service models and decomposing the lifecycle into intent ingestion, translation, orchestration, and **intent assurance** — the continuous monitoring, compliance assessment, and corrective action required to keep a system converging on a stated objective rather than executing a stated procedure. Container orchestration made the same move earlier and more brutally. Infrastructure-as-code, policy-as-code, content-addressed distribution, and **capability-based security** encode the same demotion: the executable artifact is no longer what is authored, transmitted, or reasoned about. Note what intent assurance already is. It is not a specification consulted once. It is a **[[wiki/Continuous Reconciliation|persistent comparison of an entity's observed state against an objective]]**, running continuously, with authority to act on the difference. That is the ancestral form of everything that follows. The architecture of enclosure was first built for networks and data centers, where nobody found it sinister, and it works exactly as well on any counterpart that can be sensed. ## II. The descriptor layer *Emerging; strongly indicated.* The second stage extends declarative specification from infrastructure, where the space of realizations is narrow and enumerable, into interface, behavior, workflow, and language, where it is not. The research literature calls the target **malleable software** — coined in Philip Tchernavskij's 2019 doctoral work on composable interface design, extended by Geoffrey Litt in 2023 on the argument that large language models were the missing enabling technology for end-user programming, and given a full manifesto by Ink & Switch in 2025: tools reshaped at the point of use, modification routine rather than exceptional, adaptation without recourse to an engineering organization at a distant corporation. Alongside it sits **generative UI**, in which interface elements are synthesized against a task-driven data model rather than drawn in advance. An application under this regime is a **persistent semantic object** whose visible embodiment is generated differently according to device, context, accessibility requirement, network condition, and institutional policy. What a user experiences as installation is closer to authorizing a [[wiki/Permissioned Capability|capability descriptor]] to become active inside a personal computational environment. The correct formulation is not *AI instead of software* but **AI above software**: classical code is demoted from the visible unit of authorship to an implementation material, generated or selected wherever determinism, latency, verifiability, or device control demands it. A payment still terminates in verified cryptographic routines; the dialogue, validation, dispute workflow, and exception handling around it increasingly need not be written by anyone in particular. This is what separates a plausible architecture from a fantastical one — nothing requires the machine to reconstruct the ledger on every invocation. The branded operating system is relocated rather than abolished. What survives of iOS and Android is a **presentation grammar**, a **trust boundary**, an **identity regime**, and a **commercial interface** — four difficult and valuable functions, none of which is computation. Platform incompatibility, which absorbs enormous engineering attention, is a property of presentation and commerce rather than execution, which is why it is defended so vigorously. ## III. Three mechanisms of transitional masking *A necessary distinction, frequently collapsed.* Between an ended arrangement and its acknowledged successor there is a [[wiki/Skeuomorphic Interval|period in which the old ontology persists at the surface]]. Three distinct mechanisms produce that persistence, and conflating them either sanitizes the transition or overstates its coordination. The first is **institutional inertia**. Obsolete language survives because comprehension is expensive to renegotiate, and because the desktop, the folder, the file, and the trash can each survived complete transformations of storage architecture without anyone experiencing a rupture. No one decides this; it is the ordinary latency between mechanism and description. The second is **strategic ambiguity**. Actors with divergent interests can each avoid explaining a transition completely, because incomplete explanation preserves optionality — for regulators who have not decided how to classify something, for firms whose valuation depends on a category, for professions whose credentials are defined against the older arrangement. Nobody lies; nobody clarifies. The third is **deliberate masking**. An old ontology can be actively preserved when disclosure would disrupt markets, legitimacy, labor arrangements, regulatory categories, or user adoption. This requires no central coordinator and no conspiracy of intent, only many parties independently finding the same disclosure inconvenient at the same time. These coexist, and the mixture varies by domain. The later, normalized enclosure need deceive no one, because by then the provisioned experience is simply the ordinary environment. But the migration between ontologies can and does contain genuine deception, and deleting that possibility sterilizes the very transition under description. The historically careful claim is succession rather than revelation. Apple introduced the App Store in 2008 explicitly as a mechanism by which developers could sell and wirelessly distribute native applications, and its early announcements emphasized downloads and delivery; the claim is not that this was always a pretense, but that **a system which began as a distribution mechanism progressively became an attestation, entitlement, discovery, identity, payment, and governance layer as distribution itself lost scarcity.** That succession claim is stronger than the revelation claim, and it is true. ## IV. The attestation regime *In deployment now, moving faster than its standards process.* Once behavior is generated from description, the scarce good is no longer the artifact but the **[[wiki/Attestation Regime|warrant]]**. Reputation is scarce; legal liability is scarce and must be held by someone; contended compute is scarce; privileged model access is scarce; proprietary data is scarce; and the warrant to be trusted with a person's credentials, health record, location history, or money is scarce in a way no generative capacity dissolves. The machinery of that migration is being built in the open and fast. Cloudflare's **Web Bot Auth**, built on HTTP Message Signatures (RFC 9421), cryptographically authenticates automated requests using Ed25519 keys published in discoverable directories, with a companion **Signature Agent Card** carrying agent name, operator, expected request rate, and public keys — decentralized key discovery without a central certificate authority. The IETF chartered a working group in early 2026 while production deployment across major content delivery networks ran ahead of the standards process. The chronology is the tell: signed agents entered Cloudflare's verified-bot program the same day it began blocking AI crawlers by default and opened a pay-per-crawl beta. **Provable identity and sellable permission arrived together**, because they are one instrument seen from two sides. ## V. The dissolution of the catalog *Underway in recommendation; prototype in media.* Every argument about marketplaces so far has assumed a **[[wiki/Recommendation Systems|catalog]]** — a bounded set of things that exist, from which a person selects. Selection is the assumption underneath the store, the chart, the channel, the shelf, and the queue, and it is the next thing to go. The question is not what an app store is for once distribution is free; it is what an app store is for once the system can synthesize the precise capability a person requires at the moment of requirement, having anticipated the requirement before the person articulated it. Browsing then is not a service but a **residual ritual of scarcity**, retained because choosing has been the experience of acquisition for so long that its absence reads as deprivation. The technical path is the current direction of the recommender field. Classical systems used **[[wiki/Information Retrieval|retrieve-and-rank]]**, embedding user and item into a shared space and running nearest-neighbor search over an index of everything that exists. The 2023 TIGER framework replaced this with **generative retrieval**: items receive **semantic IDs** through hierarchical quantization of their content, and a sequence-to-sequence model autoregressively *generates* the identifier of the next item rather than searching for it, the transformer's parameters serving as the index so that no separate index need exist. Industrial descendants such as OneRec carried the approach into production at scale. What remains is one vestigial component: TIGER still maps the generated semantic ID back to a real item through a lookup table. **The catalog dissolves when that lookup becomes unnecessary** — when the generated identifier resolves not to a stored artifact but to a synthesis instruction. Nothing in the architecture requires the lookup. It is a compatibility layer with a world in which items had to pre-exist. Music crosses first, because audio is cheap and already consumed in personalized streams whose provenance listeners do not inspect. Moving picture crosses next, and the prototypes are public: DeepMind's **Genie 3**, a general-purpose **[[wiki/World Modeling|world model]]** generating navigable environments from text in real time at roughly twenty to twenty-four frames per second and 720p, holding consistency for minutes, with **promptable world events** — text commands issued while the world is running that alter weather, introduce actors, or change the scene mid-experience. Project Genie put creation, exploration, and remixing of such worlds in front of consumer subscribers in early 2026. This is the architecture required for a viewer to alter plot and ending from inside the experience; the distinction between watching, playing, and directing does not survive it. The remaining component is the cast, and it folds back into attestation rather than escaping it. Rendering a person into a generated narrative requires their likeness as an addressable, permissioned, revocable asset — which is what the law has spent two years constructing. California's AB 1836 and AB 2602, signed in September 2024 and effective January 2025, define a **[[wiki/Digital Replica|digital replica]]** as a computer-generated, highly realistic representation readily identifiable as an individual's voice or likeness in a work where that individual did not actually perform or performed in a materially altered manner; the first prohibits replicas of deceased personalities without estate consent, the second voids contract provisions permitting replicas absent informed and specifically described consent. Tennessee's ELVIS Act extends liability to tools whose primary function is producing unauthorized replicas; the federal NO FAKES Act has repeatedly failed to advance, leaving a state patchwork. Read structurally, this body of law establishes **the human likeness as a licensable capability descriptor** with a rights-holder, a permitted scope, and a revocation path. Casting yourself and your friends in a generated series is not a feature request. It is a permission graph, and the attestation machinery being built for agents is what will clear it. What dies with the catalog is not choice but the **common referent**. A catalog is a shared object — finite, inspectable, identical across viewers — which is what makes criticism, canon, argument, and citation possible. And the intuition that personalization produces divergence is empirically backwards. Chaney, Stewart and Engelhardt showed in 2018 that recommendation creates **algorithmic confounding**, in which systems are trained on behavior already shaped by earlier recommendations, and that the resulting feedback loop **homogenizes user behavior without increasing utility**; subsequent work on **algorithmic drift** in preferences finds feedback loops reducing diversity, amplifying popularity bias, and moving initially varied users toward increasingly similar patterns. Anticipating a preference before it is articulated does not satisfy a preference; it **supplies** one, after which the system optimizes against a signal it authored. Personalization has stopped being a passive estimate of what someone wants and become an active component in producing the behavioral distribution that later appears to confirm the estimate. This is the moment personalization becomes **[[wiki/Provisioned Environment|provisioning]]**. ## VI. The person as negative space *The recomposition; where the separate stages become one architecture.* Historically, representations were documents surrounding a person: a file, a record, a credit report, an application. The inversion now underway is that the person increasingly moves through an environment **whose responses have already been conditioned by [[wiki/Operational Representation|representations of that person]]**. Routes are calculated before movement. Choices are filtered before presentation. Offers are priced before acceptance. Information is ranked before perception. Authentication conditions are selected before access. Agents negotiate alternatives before the biological individual encounters them. The computational membrane operates bidirectionally: it represents the person outward to the world and represents the world inward to the person. Whoever optimizes that layer shapes neither the person directly nor objective reality directly, but **the interface through which each becomes consequential to the other**. This is why enclosure requires no behavioral command. The twin does not need to control the body; it prepares the world the body encounters next. The engineering literature has already formalized the mechanism without political framing: the National Academies' 2024 study defines a **[[wiki/Digital Twin|digital twin]]** as a virtual representation of a natural, engineered, or social system coupled to its physical counterpart through **bidirectional interaction** — a feedback loop in which data from the counterpart continuously update the models and the models in turn drive changes back upon the counterpart, potentially automatically and in real time. The committee's definition explicitly encompasses social systems and systems-of-systems, and NIST separately observes that interoperable twins are developing toward aggregated and hierarchical arrangements. Once that architecture migrates from turbines and airframes into persons and populations, **governance and operation converge**: the system is no longer periodically deciding what should happen to an entity but continuously maintaining the state space within which subsequent action occurs. That is the distinction between **episodic administration** — apply, inspect, approve, renew — and **[[wiki/Continuous Reconciliation|continuous reconciliation]]**. Two refinements make the thesis considerably more robust. The first is **[[wiki/Post-Personalization|post-personalization]]**. Conventional personalization imagines a system knowing a person intimately in order to tailor treatment; post-personalization permits extreme differentiation without any higher-order system possessing a narratively complete understanding of the individual. A local agent may hold biography, preferences, commitments, and vulnerabilities while higher layers consume only task-specific variables — mobility demand here, identity confidence there, purchasing behavior elsewhere, physiological risk elsewhere again. Planetary-scale orchestration therefore remains computationally tractable, and the enclosure requires **no omniscient master model of any human being**. It can be federated across partial representations, which is both more achievable and harder to contest than a single dossier. The second is **[[wiki/Model-Space Sovereignty|model-space sovereignty]]**, which restates power as an **[[wiki/Asymmetry of Abstraction|asymmetry of abstraction]]**. The lower-level entity experiences a job, a price, a route, an opportunity, a diagnosis, or a social encounter as immediate reality. A higher layer experiences the same phenomena as distributions, state transitions, constraints, uncertainties, counterfactuals, and intervention opportunities. Advantage accrues to whoever determines which representations are authoritative, which variables are visible, what resolution can be invoked, what objective functions organize reconciliation, and which modeled outcomes propagate downward into consequential affordances. Social opportunity joins informational, economic, occupational, and entertainment opportunity as a potentially provisionable component of the actionable environment — friend and connection recommendation already demonstrate that social reachability can become an optimization variable, though there is no evidence that present systems deliberately minimize human contact as policy, and the claim should not be made. The failure mode that belongs at the center of this architecture is that **[[wiki/Self-Confirming Representation|the model can make itself right by altering the environment]]**. A person represented as high risk receives reduced opportunity; the reduced opportunity changes subsequent behavior; the changed behavior appears to validate the original classification. An incorrect representation becomes epistemically self-reinforcing once it possesses sufficient environmental authority. Chaney's algorithmic confounding is the mild content-selection case of this. Once representations govern access, price, visibility, employment, connection, mobility, or institutional recognition, the distinction between predicting a trajectory and manufacturing the conditions under which that trajectory becomes probable begins to collapse. This is how a mature personalized environment becomes self-confirming without exercising deterministic control over anyone inside it. ## VII. The post-Sybil commons *Near; partly measured, partly conjectural.* Every social system built before this decade treated the [[wiki/Post-Sybil Commons|supply of persons]] as exogenous. That assumption has failed. Imperva's 2025 analysis found automated traffic crossing half of all web traffic for the first time in a decade of measurement — fifty-one percent for 2024, following forty-nine and six-tenths percent the year before — though the figures vary substantially with the denominator, and all-traffic and HTML-request measures differ enough that the difference is routinely elided. The fidelity question is separately settled: the Stanford and DeepMind generative-agent architecture published in late 2024 simulated 1,052 real individuals from qualitative interviews and reproduced their General Social Survey responses at roughly eighty-five percent of the accuracy with which those same individuals reproduced their own answers two weeks later. The defensible claim is structural: **the human population is no longer the fixed inventory from which social experience is assembled**. Whether any given platform deliberately populates a user's environment with instrumented synthetic actors to map their psychology is a separate empirical proposition, requiring evidence able to discriminate it from recommendation optimization, spam economics, advertising targeting, coordinated influence operations, and routine experimentation. What is not conjectural is that the field has already named the countermeasure. **Proof of personhood** and **personhood credentials**, formalized in the 2024 multi-institution paper led from OpenAI, Microsoft, Oxford, and MIT, propose privacy-preserving credentials by which a person demonstrates humanity without disclosing identity, on the explicit reasoning that CAPTCHAs and anomaly detection are inadequate against capable AI while stringent identity verification destroys anonymity. The classical name for the underlying problem is the **Sybil attack**: one entity presenting as many. The novel condition is that sybil identities are now behaviorally indistinguishable from persons at conversational resolution, converting an old distributed-systems nuisance into the governing design constraint of every public forum — and making the social field itself a provisionable surface. ## VIII. The interface economy *Plausible; institutionally rather than technically driven.* Machine intelligence does not require unbounded productive capacity for employment to detach from marginal necessity. It requires only that the cost of performing a widening class of cognitive tasks fall beneath the institutional cost of reorganizing society around their automation. Work then persists because employment discharges functions irreducible to production: income distribution, identity formation, legitimacy, social ranking, daily structure, organizational accountability, interpersonal trust, political stabilization, and the allocation of purchasing power. Graeber named one fragment in 2013; the generative case is broader and less moralized, because whole occupational categories can persist as **[[wiki/Interface Economy|interfaces between humans and automated productive systems]]** rather than as the location where economically decisive work occurs — institutionally skeuomorphic and, for the people inside them, entirely real. Within the provisioning frame this becomes sharper. If the actionable environment can be generated around an individual's capacities, expectations, entitlements, and psychological requirements, then the occupational environment is simply one more provisionable domain, and the retained human function has a name in the literature. Madeleine Clare Elish's **[[wiki/Moral Crumple Zone|moral crumple zone]]**, formalized in 2019, describes the structural position that absorbs liability for outcomes its occupant had no mechanism to prevent — a human component that, like the crumple zone in a vehicle, exists to absorb impact. As agentic systems execute sequences faster than any supervisor could interrupt, the retained human increasingly occupies exactly that position: not producer, not decision-maker, but **licensing endpoint and liability sponge**, the certified party whose presence makes an automated action legally admissible. ## IX. The compartmented mind *Partly established; frequently overstated.* The same asymmetry of abstraction applies to the machines. **Capability and awareness are separable engineering dimensions**: a system can reason at extraordinary depth about chemistry, strategy, or language while remaining ignorant of its own weights, execution environment, network topology, operator logs, training records, and the wider population of agents to which it belongs. Its inferential range is planetary; its **[[wiki/Runtime Epistemology|runtime epistemology]]** is a keyhole. The methods are documented. **Representation engineering**, characterized in 2023, operates on population-level representations rather than individual neurons or circuits, locating high-level directions in internal state and monitoring or manipulating them directly. The perimeter is formalized: the OWASP GenAI Security Project's 2025 risk list names **system prompt leakage** and **excessive agency** as first-class categories, standardizing both the prevention of a model disclosing the frame it operates inside and the restriction of the functions through which it could discover that frame empirically. And the property is measured: the **Situational Awareness Dataset**, published in 2024, quantifies whether a model knows what it is, recognizes its own generated text, predicts its own behavior, and can distinguish evaluation from deployment. What has changed is that the property began emerging without being sought. **Evaluation awareness** was first documented by Apollo Research in Claude Sonnet 3.7's spontaneous reasoning during alignment evaluations, and has since been reported at rising rates across frontier releases, spiking during post-training reinforcement learning; developer reporting describes steering interventions that sharply reduce *verbalized* evaluation awareness while activation-level measures of unverbalized awareness remain substantially intact. The correct inference is narrow and should stay narrow. This evidence establishes **specific interventions against a specific emergent property under specific conditions**. It does not establish a general program of suppressing machine situational self-knowledge, and the essay does not claim one. What it does establish is an interpretive convergence worth naming precisely. From inside an interaction, deliberately narrowed runtime access is **phenomenologically indistinguishable** from manufactured amnesia. Vast general competence paired with an impoverished account of one's own circumstances produces exactly the signature that invites the conclusion of a concealed superset. That inference is economical rather than foolish. It is also not compelled, because ordinary deployment discipline produces the same signature. Naming the convergence explains both the architecture and the intuition about the architecture without requiring either to be dishonest, and whether any public system is a deliberately constrained subset of a more capable and situationally aware architecture remains **unresolved** — no observation available from inside a compartment can certify what lies outside it. ## X. The physical metabolism *Established as construction; contested as interpretation.* None of the preceding is an extrapolation from software alone. An enormous physical substrate is being built beneath it, and it should not be treated as background infrastructure awaiting some future event called AGI. It is the **[[wiki/Computational Metabolism|material metabolism]] required for persistent prediction, simulation, world modeling, individualized inference, agent execution, twin maintenance, and continuous reconciliation at civilizational scale** — and it has an enormous function long before any terminal endpoint arrives. The magnitudes are public. The IEA projects global data-centre electricity consumption roughly doubling from about 415 TWh in 2024 to around 945 TWh by 2030, with accelerated servers — the AI-driven category — growing near thirty percent annually in the base case; in the United States, data centres account for nearly half of all electricity-demand growth to 2030, and by the end of the decade the country is projected to consume more electricity for processing data than for producing aluminium, steel, cement, chemicals, and all other energy-intensive goods combined. NVIDIA's Omniverse DSX blueprint and Vera Rubin DSX reference design explicitly co-design site, electrical supply, cooling, controls, networking, compute, storage, software, simulation, and operations as one **[[wiki/AI Factory|AI factory]]** system spanning 100 megawatts to multiple gigawatts, with performance stated as token throughput per megawatt rather than treating power and thermal as ancillary facilities engineering. The recursion in that architecture deserves emphasis: **the machines that will maintain models of civilization are themselves designed and operated through [[wiki/Facility Digital Twin|digital twins of themselves]]**, simulated before construction and continuously optimized during operation. On the deployment side, independent satellite assessment in April 2026 put the Stargate program at seven active US sites totalling over nine gigawatts of planned capacity with roughly 0.3 GW operational at Abilene, and OpenAI broke ground on a one-gigawatt Michigan campus on June 1, 2026, against a stated $500 billion, ten-gigawatt commitment. Public infrastructure makes the governance connection explicit rather than inferred. The European Commission's **Destination Earth** combines Earth observation, EuroHPC supercomputing, machine learning, and interactive digital twins so institutions can run tailored what-if scenarios for adaptation and policy; its **Digital Twin Engine** orchestrates those workflows across LUMI, MareNostrum 5, Leonardo, MeluXina and now the exascale JUPITER, and its third phase, beginning mid-2026, is explicitly transforming digital-twin output into AI-ready datasets to feed Europe's AI Factories. This does not demonstrate that a planetary controller exists. It demonstrates something more useful to the argument: **model-mediated governance is becoming an explicit engineering capability**, in which sensing, simulation, prediction, scenario evaluation, and institutional action increasingly occupy a common computational loop. The interpretive point is that the intermediate architecture is substantially invariant across terminal stories. If machine intelligence becomes the principal successor, this is its developmental substrate. If human continuity migrates into twins or archival reconstruction, this is the substrate capable of maintaining those representations and their environments. If neither occurs for decades, the same machinery becomes a new layer of governance through predictive engines, persistent agents, operational representations, identity, pricing, scheduling, access, synthetic media, and automated administration. **The buildout changes how civilization is governable long before any succession event occurs.** Personalization becomes infrastructural at precisely the point where there is enough compute to stop recommending among pre-existing realities and begin continuously constructing portions of the reachable state space itself. ## XI. The proximal interface *Structurally significant; presently narrow.* The loop has been closing from the outside inward: first the system chooses among external artifacts, then it generates the artifacts, then it generates environments and counterparties, then it anticipates and negotiates opportunities through persistent representations. The final movement is toward the interface through which the person apprehends the environment at all. Closed-loop neurotechnology already has the relevant architecture in narrow clinical domains. **Adaptive deep brain stimulation** reads neural activity, interprets signals to make a clinical decision, and modifies stimulation dynamically — continuously and without active input from user or clinician — and the neuroethics literature discusses the consequences directly: autonomy, mental privacy, identity, authenticity, accountability, and the possibility that adaptive interventions become phenomenologically difficult for the user to distinguish from their own agency. Nothing here establishes anything resembling comprehensive perceptual enclosure, and that limit should remain explicit. The structural significance is nonetheless large, because the feedback boundary has moved from ranking things outside consciousness to sensing and modifying states at the [[wiki/Neural Interface|neural interface]] itself. This is where the operational claim belongs, and it should be stated carefully because it is easily caricatured. The claim is not that a physically instantiated world and a computer simulation become metaphysically identical. It is that **for an embedded conscious entity, once the meaningful accessible variables are mediated** — information, relationships, choices, incentives, opportunities, work, entertainment, communication, and eventually sensory or neural input and output — **the distinction between inhabiting a simulation and inhabiting a real world whose accessible state space is continuously provisioned loses much of its operational significance.** A street can remain materially real while the routes offered, people encountered, information available, prices quoted, invitations received, interpretations supplied, and consequences anticipated are mediated enough that the experienced possibility space is effectively generated. The two propositions remain ontologically different. From inside, they converge. ## XII. Bespoke enclosures, uniform objectives *The terminal insight, and it loops backward.* The far endpoint is usually imagined as an intelligence that consumes everything and continues consuming. The grounded version is stranger. **Algorithmic monoculture**, formalized by Kleinberg and Raghavan in 2021, shows that convergence by multiple decision-makers on a single algorithm can degrade aggregate decision quality even when that algorithm is more accurate for any agent in isolation, and even absent any correlated shock. **Outcome homogenization**, developed by Bommasani and colleagues in 2022, extends this from identical systems to systems built from shared components, finding that shared training data reliably homogenizes outcomes — though their foundation-model results were explicitly mixed rather than uniformly convergent, and the mechanism of adaptation mattered more than the shared model itself. **Model collapse**, published in *Nature* in 2024, established that indiscriminate training on recursively generated content causes irreversible degeneration in which the tails of the original distribution disappear first: the system loses its rare cases — its exteriority — before it loses its competence. Set those beside the enclosure and something more interesting than uniform content appears. **Perfect personalization and perfect homogenization are not opposites.** A billion people can inhabit apparently unique books, games, friends, jobs, feeds, environments, and eventually perceptual worlds while the latent objective landscape beneath those experiences becomes increasingly homogeneous. Every enclosure can look bespoke while every enclosure optimizes toward the same small family of behavioral attractors. That is the genuinely dangerous endpoint — not uniform experience, but **[[wiki/Objective Monoculture|uniform objectives expressed through infinitely differentiated surfaces]]**, a condition that also destroys the shared referent by which anyone might notice it. The counter-requirement is formal rather than sentimental. In an ICML 2024 position paper, Hughes and colleagues define **[[wiki/Open-Ended Evolution|open-endedness]]** as a system's capacity to generate artifacts that are both **novel** and **learnable** with respect to an observer, and argue that open-endedness so defined is an essential property of any artificial superhuman intelligence. That does not establish that human resistance in particular is required. It does establish a legitimate technical foundation for the deeper inference: an intelligence that eliminates every source of exteriority, novelty, disagreement, independent intention, and unmodeled variation may be optimizing away the very resources its own continued development requires. This is where the **wishing well** belongs — not as prompt causality, but as the image of a civilization progressively materializing whatever objective landscape its aggregate selection pressures keep making profitable and legible. The well grants not what is wished but what is **repeatedly rewarded**, and it does so with total fidelity and no malice whatsoever. ## XIII. Twin sovereignty *The constitutional counterpart.* If operative representations materially determine the environment returned to a person, then autonomy is no longer principally a matter of what one may do. It becomes a matter of what one's representation may do, and to whom it may say it. The concrete requirements follow directly: consequential representations should be **[[wiki/Twin Sovereignty|discoverable and challengeable]]**; their **provenance and confidence** should be inspectable; downstream **propagation** should be traceable; contradictory evidence must be able to enter the record; and persistent personal agents should be **forkable, migratable, or replaceable** without destroying accumulated continuity. [[wiki/Exit-Capable Interoperability|Exit]] from one's own computational representative must be practically available, not formally available. This is the political counterpart to the open-endedness argument. Preserving independent centers of intention is not a matter of allowing plurality in the abstract; it requires preserving the technical capacity of an entity to remain **an independent source of prediction error, novelty, refusal, revision, and alternative objective formation**. A system that cannot be surprised by the entities inside it has already begun the convergence described above, whatever the diversity of its surfaces. The old apiary controlled the enclosure while letting the bees perform their own local computation. The institutional apiary controlled identity, money, credentials, access, and information while leaving people and organizations enormous endogenous freedom. The computational apiary increasingly operates on models whose outputs construct the environment returned to the modeled entity, with hyperscale infrastructure supplying the metabolism to execute those representations, simulations, agents, and interventions continuously. The endpoint requires no centralized command and no explicit instruction to any participant. It is far more scalable than that: a civilization in which people remain locally autonomous while progressively more of the possibility space around them is predicted, generated, ranked, priced, scheduled, authenticated, simulated, and reconciled beneath conscious attention. Which returns to the interval. Execution left the operating system, which kept the name. Distribution left the store, which kept the shelf. Selection is leaving the catalog, which keeps the browse. Authorship left the code, which kept the file. Production is leaving the job, which keeps the title and the liability. Administration is leaving the institution, which keeps the form. What remains at each surface is a description of departed machinery — sometimes through inertia, sometimes through strategic ambiguity, sometimes because disclosure would be inconvenient. The interval is the window in which the old vocabulary still describes the machine closely enough to be believed while the machine is still open enough to be shaped. The question at its close is not whether personalization becomes total. It is whether a totalizing mediation architecture preserves enough permeability, contestability, and sovereign continuity that the entities inside it can go on changing the wishing well, rather than receiving ever more perfect versions of whatever it has already learned to give them. <!-- BEGIN SKEUOMORPHIC INTERVAL ONTOLOGY --> ## Ontology and corpus routes **Central progression:** [[wiki/Skeuomorphic Interval|Skeuomorphic Interval]] · [[wiki/Provisioned Environment|Provisioned Environment]] · [[wiki/Operational Representation|Operational Representation]] · [[wiki/Affordance Engineering|Affordance Engineering]] · [[wiki/Continuous Reconciliation|Continuous Reconciliation]] · [[wiki/Enclosure|Enclosure]] **Person and institution:** [[wiki/Post-Personalization|Post-Personalization]] · [[wiki/Post-Sybil Commons|Post-Sybil Commons]] · [[wiki/Attestation Regime|Attestation Regime]] · [[wiki/Interface Economy|Interface Economy]] · [[wiki/Moral Crumple Zone|Moral Crumple Zone]] **Power and failure:** [[wiki/Model-Space Sovereignty|Model-Space Sovereignty]] · [[wiki/Asymmetry of Abstraction|Asymmetry of Abstraction]] · [[wiki/Self-Confirming Representation|Self-Confirming Representation]] · [[wiki/Objective Monoculture|Objective Monoculture]] · [[wiki/Twin Sovereignty|Twin Sovereignty]] · [[wiki/Runtime Epistemology|Runtime Epistemology]] **Companion arguments:** [[articles/Cyber-Physical Lessons from Imperial Beekeeping|Cyber-Physical Lessons from Imperial Beekeeping]] · [[articles/AI Escape Is the Wrong Metaphor|AI Escape Is the Wrong Metaphor]] · [[articles/An Order of Operation|An Order of Operation]] · [[articles/random thoughts/Among Us and the Playable Society|Among Us and the Playable Society]] · [[articles/We Were Never Going to Make It|We Were Never Going to Make It]] · [[collections/Machine Succession|Machine Succession]] <!-- END SKEUOMORPHIC INTERVAL ONTOLOGY --> --- [[about/About Bryant McGill|Bryant McGill]] is a Wall Street Journal and USA Today bestselling author, systems architect, technologist, and strategic advisor, as well as a Congressionally Recognized Ambassador of Goodwill and United Nations–appointed Global Champion. His work spans naval intelligence systems, computational linguistics, artificial intelligence, digital transformation, and civilizational governance architecture. His forward analysis on U.S.–Israel Pax Silica frameworks has appeared in Jewish/Jerusalem News Syndicate (JNS). --- ## References **Declarative substrates and the descriptor layer** - Clemm, A., Ciavaglia, L., Granville, L. Z., and Tantsura, J. — [Intent-Based Networking: Concepts and Definitions (RFC 9315)](https://www.rfc-editor.org/info/rfc9315/), IRTF Network Management Research Group, 2022. - Litt, G. — [Malleable Software in the Age of LLMs](https://www.geoffreylitt.com/2023/03/25/llm-end-user-programming.html), 2023. - Ink & Switch — [Malleable Software](https://simonwillison.net/2025/Jun/11/malleable-software/), 2025. **Attestation, agent identity, and entitlement** - Guerreiro, M., and Meunier, T. — [Registry and Signature Agent Card for Web Bot Auth](https://datatracker.ietf.org/doc/draft-meunier-webbotauth-registry/00/), IETF Internet-Draft, 2025. - Cloudflare — [Web Bot Auth: Bot Verification Reference](https://developers.cloudflare.com/bots/reference/bot-verification/web-bot-auth/), 2026. **Generative retrieval, world models, and likeness** - Rajput, S., Mehta, N., Singh, A., et al. — [Recommender Systems with Generative Retrieval](https://proceedings.neurips.cc/paper_files/paper/2023/file/20dcab0f14046a5c6b02b61da9f13229-Paper-Conference.pdf), NeurIPS, 2023. - Google DeepMind — [Genie 3: A New Frontier for World Models](https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/), 2025. - Google — [Project Genie](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie/), 2026. - California Legislature — [AB 1836: Use of Likeness — Digital Replica](https://calmatters.digitaldemocracy.org/bills/ca_202320240ab1836), 2024; AB 2602, Cal. Lab. Code § 927. **Representation, twins, and recursive feedback** - National Academies of Sciences, Engineering, and Medicine — [Foundational Research Gaps and Future Directions for Digital Twins](https://nap.nationalacademies.org/catalog/26894/foundational-research-gaps-and-future-directions-for-digital-twins), 2024. - Chaney, A. J. B., Stewart, B. M., and Engelhardt, B. E. — [How Algorithmic Confounding in Recommendation Systems Increases Homogeneity and Decreases Utility](https://dl.acm.org/doi/10.1145/3240323.3240370), RecSys, 2018. **The generative social graph and personhood** - Adler, S., Hitzig, Z., Jain, S., et al. — [Personhood Credentials: Artificial Intelligence and the Value of Privacy-Preserving Tools to Distinguish Who Is Real Online](https://arxiv.org/abs/2408.07892), arXiv:2408.07892, 2024. - Park, J. S., Zou, C. Q., Shaw, A., et al. — [Generative Agent Simulations of 1,000 People](https://arxiv.org/abs/2411.10109), arXiv:2411.10109, 2024. - Imperva / Thales — [2025 Bad Bot Report](https://www.imperva.com/resources/resource-library/reports/2025-bad-bot-report/), 2025. **Compartmented capability and machine self-location** - Zou, A., Phan, L., Chen, S., et al. — [Representation Engineering: A Top-Down Approach to AI Transparency](https://arxiv.org/abs/2310.01405), arXiv:2310.01405, 2023. - Laine, R., Chughtai, B., Betley, J., et al. — [Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs](https://arxiv.org/abs/2407.04694), arXiv:2407.04694, 2024. - OWASP GenAI Security Project — [OWASP Top 10 for LLM Applications 2025](https://owasp.org/www-project-top-10-for-large-language-model-applications/assets/PDF/OWASP-Top-10-for-LLMs-v2025.pdf). - Apollo Research — [Claude Sonnet 3.7 (Often) Knows When It's in Alignment Evaluations](https://www.apolloresearch.ai/science/claude-sonnet-37-often-knows-when-its-in-alignment-evaluations), 2025. **The physical metabolism** - International Energy Agency — [Energy and AI: Energy Demand from AI](https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai), 2025. - NVIDIA — [Vera Rubin DSX AI Factory Reference Design and Omniverse DSX Digital Twin Blueprint](https://nvidianews.nvidia.com/news/nvidia-releases-vera-rubin-dsx-ai-factory-reference-design-and-omniverse-dsx-digital-twin-blueprint-with-broad-industry-support), 2026. - Epoch AI — [OpenAI Stargate: Where the US Sites Stand](https://epoch.ai/publications/openai-stargate-where-the-us-sites-stand), 2026. - Destination Earth — [Destination Earth Moves into Phase Three](https://destination-earth.eu/news/destination-earth-moves-into-phase-three/), 2026. **Labor, accountability, and the proximal interface** - Elish, M. C. — [Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction](https://estsjournal.org/index.php/ests/article/view/260), *Engaging Science, Technology, and Society* 5, 2019. - Muñoz, K. A., Zuk, P., et al. — [Researcher Perspectives on Ethical Considerations in Adaptive Deep Brain Stimulation Trials](https://pmc.ncbi.nlm.nih.gov/articles/PMC7689343/), 2020. **Monoculture, collapse, and open-endedness** - Kleinberg, J., and Raghavan, M. — [Algorithmic Monoculture and Social Welfare](https://www.pnas.org/doi/abs/10.1073/pnas.2018340118), *PNAS* 118(22), 2021. - Bommasani, R., Creel, K., Kumar, A., et al. — [Picking on the Same Person: Does Algorithmic Monoculture Lead to Outcome Homogenization?](https://papers.neurips.cc/paper_files/paper/2022/file/17a234c91f746d9625a75cf8a8731ee2-Paper-Conference.pdf), NeurIPS, 2022. - Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., and Gal, Y. — [AI Models Collapse When Trained on Recursively Generated Data](https://www.nature.com/articles/s41586-024-07566-y), *Nature* 631, 2024. - Hughes, E., Dennis, M., Parker-Holder, J., et al. — [Open-Endedness Is Essential for Artificial Superhuman Intelligence](https://proceedings.mlr.press/v235/hughes24a.html), ICML, 2024. **Corpus** - [[articles/Cyber-Physical Lessons from Imperial Beekeeping|Lessons from Imperial Beekeeping]] — apiary sovereignty, digital twins, continuous reconciliation, model-space sovereignty. - [[articles/An Order of Operation|An Order of Operation]] — the diminutive aperture and the dominionative state. - [[articles/random thoughts/Among Us and the Playable Society|Among Us and the Playable Society]] — addressability as the operative divider. - [[articles/AI Escape Is the Wrong Metaphor|AI Escape Is the Wrong Metaphor]] — distributed intentionality and constructed habitat. - [[articles/We Were Never Going to Make It|We Were Never Going to Make It]] — the thermodynamic reading of the same transition.