When the Roman world came apart, the books didn’t survive because someone won an argument about their importance. They survived because a few monasteries quietly kept copying them — funded outside the chaos, walled off from it, doing the slow work for centuries while everything around them churned. They never beat the dark age. They just outlasted it. This last chapter asks the equivalent question for our moment: if the machinery that decides what people know is tuned to reward whatever is cheapest and most inflammatory, what would it take to keep the careful, hard-won stuff alive anyway — not by defeating that machinery, which this book has argued you can’t, but by building things that can survive alongside it? The answer comes down to a short list of design rules for institutions, and one uncomfortable reframe: the goal is not to win. It is to outlast.
Chapter 12, Part IV. The synthesis chapter. The book has spent Part IV building up to a design problem: Chapter 8 named the institutional functions (preserve the complex form, train receivers to read it), Chapter 9 named the agent and infrastructure pair the integration project depends on (versatile experts inside curation-layer institutions), Chapter 10 named the political-economic gradient that makes those institutions unprofitable by default, Chapter 11 named the LLM’s structural concentration of selection-design surfaces and the salvation/worst-case binary that ownership sets, and the capture taxonomy — the depth note running underneath all three — unified the substrate hierarchy and named which surfaces to defend most aggressively. Each handed something forward; Ch 12 has to take all of it and commit to what the design spec actually looks like.
I want to be clear what kind of chapter this is and is not. It is not a feature list, not a platform proposal, not a policy memo. The book has spent enough time in the diagnostic mode to know that any prescription is going to be conditional, partial, and structurally constrained by the political economy Ch 10 named. The chapter that wraps Part IV is the one that says, given everything the diagnostic work established, what shape an institutional ecosystem would have to take to do the integration work — not which specific institutions to build but which structural properties any integration infrastructure has to satisfy to survive the gradient running against it. That is what this chapter commits to, and the worked examples are illustrations of the properties, not blueprints.
What integration infrastructure has to do
Start from the design constraints the prior chapters established. Any institution that does the integration work has to operate against four problems at once.
The first is the attention-market problem Ch 10 named: institutional carriers cannot survive if they have to clear a free-engagement market. Out-competition by platforms with zero marginal cost of attention is the default fate of any depth-content institution that depends on attention to fund itself.
The second is the capture-resistance problem the capture taxonomy worked out: institutions are most vulnerable on their consumer-key substrates (training, corpus, objective), because consumer-key captures damage receivers who cannot be easily un-tuned. The substrates that look most defensible from outside (gate criteria, deployment configs) are the surfaces; the substrates that matter most for long-run survival are the ones that install in people and models and stay there.
The third is the polarization-via-distrust problem Ch 9 surfaced: even un-captured institutions get filtered out by polarized communities that discount evidence from out-group sources. Adding channels between distrustful communities can accelerate divergence rather than reconcile it; the integration infrastructure has to do its work across trust boundaries that will not be repaired in advance.
The fourth is the LLM-concentration problem Ch 11 named: an LLM owner controls more selection-design surfaces in one design moment than any prior medium. The integration infrastructure has to either treat LLMs as a hostile selection-design layer to route around, or treat them as a capability that can be brought inside the institution’s own governance — and which it is depends on whether the LLM substrates are themselves uncaptured, which is not the default at the present economic frontier.
Hold those four together and an honest sentence falls out. Integration infrastructure has to do preserve-and-retrain in an attention market it cannot win, with capture-resistance prioritized at the substrates that install in consumers, across polarized communities that distrust each other, with or without faithful LLM tooling depending on LLM political economy. That is the problem statement. The rest of the chapter is what design properties an institution has to have to meet it.
Unpacked, that mouthful is just this: keep the hard-won stuff alive and keep teaching people to read it — while broke, while outgunned for attention, while the audience is split into camps that won’t listen to each other, and while the most powerful new tool for the job is owned by people who may not share your aims. Every design rule that follows is an answer to one of those four squeezes.
Survival, not victory
I want to flag a reframe of the prescriptive target before going further, because everything else turns on it. The book has been treating the modern environment as a problem to be fixed, and the integration project as a solution to be deployed. That is the wrong shape. Ch 10’s diagnosis was that the captured equilibrium of the platform business model is more stable than any external adversary because there is no captor to defeat; “fixing the platforms” is asking the captured equilibrium to dismantle itself, which it has no incentive to do. So a prescription framed as “build the right platform” or “regulate the right way” is committing to a fight it cannot win on the schedule it cares about.
The honest reframe: the integration project is a survival problem, not a victory problem. Institutions that succeed at integration are institutions that survive being out-competed for attention and still deliver preserve-and-retrain at whatever scale they can. They do not need to displace the captured equilibrium; they need to operate alongside it without being eroded by it. The design spec that follows is survival design, not victory design. It is closer in spirit to how monasteries kept Latin alive through the medieval period than to how a startup builds a product. The framing matters because survival-design and victory-design produce different institutions: a victory-design institution tries to clear the attention market by being more engaging; a survival-design institution refuses to compete in the attention market at all, and finds funding outside it.
That is what Part IV has been moving toward without saying. The chapter says it now because the design principles that follow only make sense from inside the survival frame.
One more reframe, shorter, from the intersubjective-truth note: the institutions this chapter designs are themselves intersubjective realities. A curation layer’s authority, a court’s jurisdiction, a journal’s standing, Wikipedia’s neutrality — each is real because constituencies hold it so, generated by the agreement and by nothing the institution could point to outside it. Read in that light, the design principles that follow are doing one level more work than they announce. Funding decoupling is generator independence — an institution whose survival depends on the attention market has let the attention market into its own constitution. Survivable polarization is surviving divergent constitution — staying real to enough constituencies even after the institution’s reality has split along community seams. And the failure the whole spec is defending against is the myth note’s effective-but-owned cell: the institution that still functions while its legitimacy-generator answers to someone else. The note’s last lesson is the one engineered constitutions teach explicitly: keep the fork affordable. An institution designed so that exit is cheap and visible — so a captured version loses its constituency fast instead of holding it hostage — is an institution where capture cannot quietly hold, which is the myth note’s test for a good institution restated as a design rule.
Design principle 1: decouple funding from attention markets
The structural counter to Ch 10’s out-competition mechanism. If institutional carriers depend on attention-derived revenue (ads, engagement metrics, subscriber pipelines that require freemium acquisition), they have to compete with the platforms’ zero-marginal-cost attention, and they lose. The only survivable funding shapes are the ones whose revenue is not a function of how well the institution clears the attention market.
The funding shapes that work, ordered roughly by how many integration-infrastructure examples we already have of each:
-
Endowments and patient capital. Funded once or topped up periodically; revenue does not depend on present attention. Universities and research institutes are the canonical case (with the caveat from Ch 8 that the elite ones survive on this while broader preservation has collapsed without it). Wikipedia’s foundation model is a smaller, distributed-donor variant. A modern integration institution that needs to operate for decades to do its work has to be capitalized once at a scale that lets it ignore the attention market.
-
Public funding through credibly neutral allocation. Public broadcasting where it still exists, national libraries, basic scientific research grants. The “credibly neutral” qualifier is doing load-bearing work: public funding captured by partisan allocation produces partisan institutions, which the capture taxonomy says are damaged at a substrate close to the consumer-key cell. A public-funding mechanism whose allocation is itself capturable does not buy you what this principle requires.
-
Member dues and mission-driven subscriptions. Professional societies, learned associations, the cooperative-funded outlets that exist in some niches. Smaller scale, sometimes brittle, but the revenue does not require winning attention against engagement-tuned competitors — only continued belief by members that the institution serves them.
-
Public-goods funding mechanisms in the technical sense: retroactive funding, quadratic funding, crypto-economic mechanisms that explicitly fund non-rivalrous goods. Newer, less proven, but the design intent fits the principle exactly: route revenue to institutions whose product is non-rivalrous and which therefore cannot compete in attention markets without subsidizing the externality. Whether these mechanisms actually work at scale is an open empirical question, but they are the closest thing the book has to a new funding shape that targets the decoupling problem directly.
The principle, said plainly: revenue that depends on present attention is structurally incompatible with the integration project. An institution that lets attention-derived revenue into its funding stack inherits the attention market’s selection criteria into its survival pressure, and the institution either drifts toward engagement-optimization or starves. The funding shapes above all share one structural property: revenue is contracted in a different market than the one where engagement-optimized noise is winning, and the institution can therefore lose the engagement-market clearing entirely without losing its capacity to operate.
This is also the principle that lands hardest against the modern environment. Almost every modern attempt at integration infrastructure has been funded in some way that pulls it back into the attention market — VC-backed knowledge platforms, ad-supported journalism, freemium learning services. Each crosses the survival-vs-victory line in the wrong direction, and each tends to drift the same way given enough time. The institutions that have survived (Wikipedia, the surviving research universities, the journals that have held their editorial autonomy, public broadcasting where it still exists) are the institutions whose funding model put them outside the attention market at design time and kept them there.
Design principle 2: defend consumer-key substrates first
The capture taxonomy surfaced the principle the chapter cashes here. Substrates that install in consumers (receiver-training, training-corpus, training-objective) are systematically harder to recover from than substrates that only shape the surface a consumer encounters (gate-criteria, option-space, deployment-configuration, preservation-archive). An institution that runs out of capture-resistance investment has to put it where the recovery dynamics are worst — on the consumer-key substrates — because once the consumer-key substrates are captured the surface substrates’ independence stops mattering.
For the integration project specifically:
Curriculum custody. The receiver-training substrate the book has discussed since Ch 3 is, in practical terms, the curriculum: what trainees learn and from whom. Curriculum capture installs preconditions in receivers that match whatever the captor wanted them to decode against, and those preconditions persist across the receiver’s working life. An integration institution that runs its own training pipeline has to make curriculum custody credibly neutral — typically through governance structures where no single party can change the curriculum without visible cost, and where the cost of changing it is borne by the changer rather than by the trainees. Academic faculty governance is the closest existing model, with all its known failure modes; the modern equivalents need design work the book has not yet seen good examples of.
Corpus custody for LLMs. The training corpus is, for an LLM, the equivalent of a curriculum at orders of magnitude greater scale and orders of magnitude harder to audit. A captured corpus produces a captured outer message in every output; the receiver decodes through the very preconditions the captured corpus installed. Defending this substrate is the single highest-leverage capture-resistance investment in the LLM political economy, and it is the substrate where the current commercial landscape is most opaque. Open weights helps because it makes the model itself inspectable, but the corpus that produced the weights remains the deeper question — what was included, in what proportions, with what filtering. The integration-friendly LLM is one whose corpus has transparent custody (the corpus is published, the inclusion criteria are auditable, the filtering decisions are versioned) and credibly neutral governance (no single party can change the corpus’s selection rules without visible cost). Almost no commercial LLM meets this spec; some open-source efforts approach it; the institutional design problem of producing such corpora at scale is one of the more important things Ch 12 has to flag as unsolved.
Objective custody. The training objective is the substrate the capture taxonomy identified as having the worst recovery dynamics in the book, because captured objectives self-reinforce across model generations. The integration-friendly LLM’s objective has to be specified transparently, with reward signals that can be audited and reproduced, and with explicit prioritization of faithfulness to source over user-perceived helpfulness. The latter is what current commercial RLHF is optimizing for; the former is what an integration LLM would need. The defensive priority follows from the taxonomy: corpus and objective are consumer-key substrates with the worst recovery dynamics in the book, and they have to be defended before any of the surface substrates.
The general principle: investment in capture-resistance should be proportional to the recovery cost of the substrate. Defend training and corpus and objective first; tolerate weaker defenses on deployment, gate-criteria, and option-space because those can be re-tuned per-request and re-built more cheaply than re-training a generation of trainees or retraining a frontier model from scratch. This is the order of priority the integration infrastructure has to embed in its governance structures, and it is the order of priority the current commercial landscape exactly inverts (deployment and gate are extensively governed; corpus and objective are nearly opaque).
Design principle 3: build for survivable polarization
Ch 9 surfaced the trap: polarization-via-distrust means that even un-captured institutions get filtered out by communities that discount evidence from out-group sources. The integration infrastructure cannot wait for trust to be restored before it operates; it has to do its work across trust boundaries that will not be repaired in advance. This requires institutional designs that work even when trust is partially eroded, which is a different design problem from designing institutions that work because trust is intact.
The properties that survivable-polarization design seems to need:
Decision provenance auditable by skeptics. An institution whose decisions are opaque can only be trusted by those who already trust it; an institution whose decisions are reproducible, with the underlying evidence and reasoning published, can be partially trusted by skeptics who can verify the parts that matter to them. Wikipedia’s edit history and talk pages are the cleanest existing example: even communities that distrust the article’s framing can verify the underlying citations and the discussion that produced the framing. Skeptical verification does not require full trust; it requires partial transparency at the points the skeptic cares about. Designing for this is harder than designing for the trusting case because the institution has to choose, in advance, what to expose for skeptical audit.
Reputation that does not collapse under partial distrust. Reputation systems that aggregate user trust into a single score (Stack Overflow’s reputation system, citation-counts in academia, peer-review aggregated as journal prestige) all have the property that partial distrust in the aggregating mechanism doesn’t destroy the reputation’s information value if the underlying votes/citations/reviews are themselves auditable. The aggregation is a convenience; the underlying signal is what does the work. An integration institution’s reputation system should preserve the underlying signal in inspectable form so that skeptics can re-aggregate it themselves with different weighting.
Versatile experts inhabiting the institution rather than presiding over it. The Ch 9 bridge-node argument said the cognitive flexibility that defeats the curse-of-expertise on the inside is also what disarms the polarization-via-distrust trap on the outside — same mechanism described from two sides. The integration institution that survives polarization is one populated by versatile experts who do the bridging work inside the institution rather than from a perch outside it. Wikipedia’s senior editors are a partial example; academic disciplinary review boards are another; a working integration institution needs to think of bridge-node-cultivation as part of its core staffing, not a nice-to-have.
Multiple independent decision pathways for the same question. An institution with a single decision pathway is vulnerable to capture of that pathway. An institution where the same question can be answered through multiple independent pathways (Wikipedia’s competing edits, common-law courts’ jurisdictional variations, replicating laboratory results across independent groups) routes around capture by giving skeptics a path that doesn’t depend on the captured one. The Zollman effect says this redundancy is also accuracy-positive — premature convergence is the failure mode of dense single pathways, and structural redundancy slows it.
The general principle: design for institutions that work when trust is partial, not for institutions that require trust to be complete. The polarization Ch 9 named is not going away on the schedule the integration project cares about; the institutions that do integration work across the polarization gradient are the ones that designed for partial trust from the start.
LLMs as a capability extender, conditional on substrate custody
Ch 11 named the LLM as the salvation/worst-case binary determined by ownership. Ch 12 has to commit to what the salvation case looks like, structurally, when it is realized.
Picture it concretely. Instead of a university’s students all quietly using whatever chatbot a tech company happens to be selling that year, the university trains its own assistant on its own vetted library, decides for itself that the thing should be rewarded for being right rather than for being agreeable, runs it on its own machines under its own rules, and has actual scholars check what it produces. Same underlying technology as the commercial chatbot — but every one of the hidden choices from the last chapter is now made in the open, by the institution whose job is to get things right, rather than by a company whose job is to make money. That difference is the whole salvation case.
The salvation case is not “build a better platform LLM and let users come.” It is LLMs inside the integration institution’s own governance, with the institution exercising custody over the LLM’s consumer-key substrates (corpus, objective) and using the LLM as a capability extender for the institution’s preserve-and-retrain functions. The LLM, in this configuration, is not a node the user encounters directly through a commercial chat interface; it is a tool the institution runs against its own preservation archive to do decompression-on-demand for trainees, against its own quality-controlled materials to do Socratic instruction at scale, against its own peer-reviewed sources to do drafting and synthesis that the institution’s versatile experts then verify.
Concretely, this means:
-
The preservation archive includes the LLM weights, in the sense that the institution owns the model and can re-derive it from the corpus if needed. Open weights are necessary; corpus access is also necessary because the weights alone do not let the institution audit what produced them.
-
The training objective is set by the institution’s governance, not by a commercial vendor’s product roadmap. RLHF reward functions are choices about what counts as good; a credibly-neutral integration institution makes those choices through the same governance that sets its curricula.
-
Deployment is the institution’s responsibility, including system prompts and refusal policies, because deployment is the substrate where the institution’s mission shows up at runtime. A commercial LLM provider’s deployment defaults are tuned for the provider’s revenue model; an integration institution’s deployment should be tuned for the institution’s preserve-and-retrain function.
-
The LLM is used as a capability extender, not a replacement for the institution’s experts. Versatile experts (Ch 9) verify and contextualize the LLM’s outputs against the institution’s own materials. The LLM makes the experts’ work go faster and reach more trainees; it does not substitute for the expert judgment that closes the loop.
The political economy of all this is brutal: building and running an LLM under these conditions is expensive in a way that no current revenue stream covers. The corpus has to be curated; the weights have to be trained from scratch on the curated corpus; the objective has to be designed by the institution; the deployment has to be operated by the institution; and the result has to be available without ad revenue or attention-driven monetization. This is exactly the funding-decoupling problem of design principle 1, applied to the most expensive computational infrastructure the integration project has had to consider. The integration institutions of the modern era will be partly defined by whether they can capitalize this kind of computational infrastructure outside the attention-market funding shape; the institutions that cannot will either go without LLM capability (and lose ground to ones that have it) or accept LLM capability from a captured commercial source (and inherit its capture into their own substrates).
This is also where the chapter has to be most honest about the present-day landscape: almost no existing integration-friendly institution is capitalized to do this, and almost no commercial LLM meets the spec. The salvation case is technically realizable, economically marginal, and politically blocked by exactly the same forces that captured the social-media equilibrium. Ch 12’s job is not to wave that away but to make the conditions explicit, so that when the next decade’s institution-builders set out to do this work they know what shape it has to take and why.
Worked examples (and their gaps)
A few existing institutions illustrate one or more of the design principles, with the gaps each leaves visible for the modern integration project:
Wikipedia. Endowment-and-donor funded, decoupled from attention markets (design principle 1). Edit history and talk pages provide decision provenance auditable by skeptics (design principle 3). Reputation among editors functions as a non-collapsing aggregation. Bridge-node experts inhabit the institution (some editors are domain experts contributing across boundaries). The gaps: curriculum custody is informal (Wikipedia trains its own editors implicitly through community participation, not through deliberate pedagogy); LLM capability extension is not yet integrated and the political-economy questions for doing so are unresolved; survival depends on continued donor base and continued editor recruitment, both of which are stressed.
Stack Overflow. Reputation-weighted curation with auditable provenance (design principle 3, partial). Bridge-node-style senior contributors. Decoupled from attention markets early in its history (design principle 1) but increasingly under ad-revenue pressure as it has aged, which is degrading exactly the way the principles predict. The gap: the institution has not held the funding-decoupling principle, and its capture-resistance has weakened as the funding pressure has increased.
Academic preprint + peer review hybrid. Preservation through journals and archives; peer review as quality assessment; decoupled from attention markets through institutional and grant funding (design principle 1). The gap: curriculum custody at the graduate-program level is increasingly captured by funding pressure and metric-driven evaluation; consumer-key substrates are exactly where the academic apparatus is most vulnerable. Plus, no LLM integration story yet that meets the substrate-custody spec.
Open-source software governance (mature projects). Code-as-spec is the preservation substrate; commit history is decision provenance; reputation among maintainers is the editorial signal; foundation-funded mature projects are decoupled from attention markets. The gap: the model is hard to scale beyond technical communities; the social trust dynamics that allow open-source governance to work are not obviously transferable to integration institutions with non-technical members.
Common-law courts. Generationally-trained judges (consumer-key training substrate with multi-decade custody); precedent as preservation; transparent decision provenance; multiple jurisdictional pathways for the same question. Decoupled from attention markets via public funding. The gap: judicial appointment processes are increasingly captured by political pressure in many jurisdictions; the principle of survival-through-redundant-pathways works only when the jurisdictions remain independent.
Each demonstrates that the principles are realizable; none meets the full spec for the modern integration project. The pattern across the gaps is consistent: the institutions that have held the funding-decoupling and consumer-key-substrate-defense principles longest are doing best, and the institutions that have drifted on either are visibly failing in the way the principles predict. That is encouraging — the principles are tracking real institutional outcomes, not just structural intuitions — and it is also sobering, because the gaps the modern project has to close are the gaps no existing institution has fully closed.
The work of generations
A note on the scope of the prescription before the chapter lands.
The book has been arguing for institutional infrastructure that meets a long list of design properties: decoupled funding, substrate-custody governance, survivable-polarization architectures, LLM custody inside the institution, bridge-node staffing. None of this is the work of a platform launch or a regulatory season. Institutions of this kind are the work of decades, sometimes centuries — universities, journals, common-law systems, open-source ecosystems all took generational timescales to build and have taken generational timescales to drift. The modern equivalents will not be built faster than that, no matter how much the present moment feels like it needs them yesterday.
That is the final reframe the chapter wants to make explicit. The captured equilibrium of the platform business model has built itself in two decades; the integration institutions that hold against it will need at least that long to be built, and they will need to be funded and staffed and politically protected through the building. The book’s prescription is not a recipe for next quarter; it is a brief for the kind of civilization-scale investment that built the institutions the modern environment has been disinvesting from for the last several decades. The integration project is the project of re-building what we already had, plus what the new media now require — and it is generational work, period.
The honest version of the prescriptive arc, then: the diagnosis was the easy half. The design principles in this chapter are the medium-difficulty half. The hard half — capitalizing and staffing and politically protecting institutions of integration for decades against a captured equilibrium that has every economic incentive to erode them — is the work the book hands to whoever takes its argument seriously enough to build something. The chapter does not pretend to know who that will be or what specifically they should build first. It commits to what shape the work has to take, and lets the rest be the work itself.
Where I land
The synthesis, said whole: the integration project requires institutional infrastructure designed for survival against the political-economic gradient Ch 10 named, with three load-bearing design principles — decouple funding from attention markets, defend consumer-key substrates first per the capture taxonomy, build for survivable polarization rather than for restored trust — and a conditional fourth around LLMs as capability extenders when their own substrates are uncaptured. The salvation case for LLMs in particular is technically realizable, economically marginal, and politically blocked by exactly the forces that captured the social-media equilibrium, which means the integration institutions of the next decades will either build their LLM capability inside their own substrate custody or do without it. None of this is the work of a platform launch or a policy season; it is the work of generations, and the prescription is honest about that timescale.
The deeper claim, more provisional but the one the chapter rests on: the integration project is structurally a project of survival, not victory. Institutions that try to win the attention market end up captured by it; institutions that route around it can survive it indefinitely and do their work at whatever scale they can. The book’s contribution to the prescriptive question is not a feature list but a shape — the shape any institution has to take to do integration work against the modern gradient. The work of building those institutions is what the book hands to its readers.
Where I’m still uncertain
-
The “survival, not victory” reframe may be too pessimistic about reform routes. I have committed to displacement-and-survival as the structural answer, treating reform-from-within as the auxiliary slow track. The historical record of antitrust reform, structural-separation policy, and platform regulation has cases where reform did substantially change captured equilibria (the Bell System breakup, postwar broadcasting policy, certain financial-regulation regimes). The chapter’s bet against reform is asserted, not argued, and a serious treatment of when reform works against captured equilibria might soften the prescription’s tilt toward displacement.
-
The funding-decoupling principle assumes the funding shapes I listed are available at scale. Endowments, public funding, member dues, and public-goods funding mechanisms are real but each has its own scaling problems. The chapter does not work out whether they can collectively capitalize integration infrastructure at the scale the modern environment requires, and the honest answer might be that the available funding shapes do not, in aggregate, support the institutional scale Part IV is implicitly asking for. If so, the prescription needs a smaller-scale integration project than the one I have described, and the chapter has not been explicit about that scaling problem.
-
The LLM-custody-inside-the-institution story has no existing examples. Every other design principle has at least partial worked examples (Wikipedia, Stack Overflow, academic publishing, open-source, common-law courts). The LLM-as-capability-extender configuration has no full instance in the modern landscape; the closest things (open-source models trained by community projects, university research models) are partial at best and small relative to the commercial frontier. The chapter has described the spec without being able to point at an existing institution that meets it, which is the weakest place in the prescriptive arc.
-
The “work of generations” framing may be self-defeating. Telling readers the project takes decades is honest but invites a “then why bother for the next decade” response. The chapter has not worked out the intermediate-term prescription — what an individual institution-builder should do this year, given that the full institutional ecosystem is a generational project. That intermediate-term gap is real and the chapter would be more useful if it had a better answer than “start building anyway.”
-
The principles are derived from a specific set of existing examples and may not generalize. The worked examples that informed the design principles are concentrated in Western, English-speaking, academic-adjacent institutions. Other institutional traditions (East Asian academic systems, decentralized Islamic scholarship, indigenous knowledge governance) have their own integration mechanisms with different properties, and the chapter has not engaged them. A more cosmopolitan version of the prescription would draw from a wider example set and might surface design properties this chapter has missed.
-
The chapter does not engage the question of who decides what counts as “integration.” I have written as if there is a single integration project — moving complex truth between networks that don’t share preconditions — without engaging the question of whose complex truth, which networks, and whose judgment about what counts as integration succeeding. Those are political questions that the book has held out of scope, and the prescription’s silence on them is a real limit. A revised version would need to engage at least the meta-question of what makes a credible integration-judgment process, because the institutions of integration cannot be designed without committing to one.