Cross-Pollination, Not Transplantation

The previous inquiry ended at the point where comparison becomes insufficient. I could now imagine a political system asking a much better question than which foreign model it should copy. It could ask which mechanism had solved a similar problem elsewhere, what conditions had kept that mechanism alive, where it had failed, and which part might travel. But knowing what might travel does not tell anyone what should be planted here.

That gap had seemed small to me while I was studying institutions from a distance. Once I tried to cross it, the gap widened. A mechanism arrives in a society that already has laws, offices, rights, habits, vetoes, professional roles, fiscal limits, political loyalties, and memories of earlier reform. If it is inserted without changing any of those relations, it may not work. If it is changed too freely, the very relation that made it useful may disappear. Adaptation is therefore not the easy step after comparison. It is the point at which political learning becomes political authorship.

On Luminara, we were slow to understand this because our borrowing was usually narrated from the donor’s side. A powerful society possessed an institution; another society lacked it; reform moved the visible form across the boundary. The receiving society then appeared in the history mainly as the place where the model succeeded, resisted, or failed. Even when the transfer was voluntary, the story made the donor the inventor and the recipient the implementation site.

Earth had already shown me the flaw in that story. Sweden’s ombudsman did not remain Swedish when it entered New Zealand or South Africa; each political ecology changed the office’s jurisdiction, consequence, and relation to other institutions. Yet my comparison still stopped before the most difficult question. What would it mean to design the change deliberately—to take a useful relation from elsewhere and rebuild it so that the receiving society remained the author of its own political development?

The first case I chose looked almost designed to produce confusion, because its name has long been associated with political command: the plan.

Planning Without Command

France emerged from the Second World War with damaged infrastructure, shortages, an urgent need to modernize industry, and too many interdependent investment decisions to leave each sector to discover the national bottleneck separately. In 1946 the government created the Commissariat général du Plan under Jean Monnet and adopted the first modernization and equipment plan. Coal, electricity, transport, steel, cement, agricultural machinery, and other strategic capacities had to be expanded in combinations; one shortage could make investment in another sector useless.

The word planning invites a familiar political inference. If a government identifies priorities across the economy, coordinates scarce resources, and keeps a multi-year horizon, perhaps concentrated political authority is part of the mechanism. France’s own institutional history makes that inference difficult. France Stratégie traces a direct intellectual and practical line from Monnet’s wartime experience with American production-coordination boards to the early work of the French planning commission. The ancestor was coordination under emergency. The postwar descendant was deliberately rebuilt for a different political relation.

The difference was not cosmetic. The French Plan was not Soviet Gosplan translated into a parliamentary language. Its commissions brought together administrators, employers, trade unions, experts, and sectoral participants to develop shared diagnoses and priorities. Contemporary OECD accounts described tripartite planning commissions that did not set authoritative production targets for individual firms. The government retained responsibility for economic policy; annual budgets remained acts of government and Parliament; the plan supplied a longer horizon within which investment decisions, public priorities, and private expectations could be coordinated.

This was not laissez-faire with a committee attached. The French state owned major enterprises, controlled important financial levers, directed public investment, and could make the plan consequential through budgets and policy. Calling the plan indicative therefore does not mean that it lacked power. It means that the mechanism cannot be reduced to one command relation in which a central authority dictates every contribution. Consultation, public investment, information, expectation, and political authorization were combined to make long-horizon coordination possible without turning the plan itself into a sovereign economic office.

That distinction mattered to me because the first source of cross-pollination was not a finished institution but a change in relation. Wartime boards had helped coordinate production under emergency conditions in which unusually concentrated direction was used. France preserved the capacity to see interdependence and sequence investment, but rewrote the authority around it for peacetime. Employers and unions entered the diagnostic process; ministries and public enterprises carried implementation; Parliament remained part of the political route; firms did not become administrative subdivisions of the plan. The capability survived by losing some of the authority with which its antecedent had operated.

Norway had already cost me the contrast between democratic short-termism and authoritarian long-termism. France shows something the Norwegian case did not: that the coordinating capacity itself can be rebuilt under a different authority rather than merely disciplined by one. A political system does not acquire a long horizon by becoming less democratic in the abstract. It acquires one by creating institutions that can carry information, commitments, and coordination farther into the future than the ordinary political cycle while leaving elected authority able to revise the course. The French arrangement was one historical solution to that problem, not the solution.

Its later history is part of the lesson. The early plans were made under reconstruction conditions in which bottlenecks were unusually visible and public control over investment was unusually strong. As the economy became more open and complex, later plans became more indicative and their forecasts less reliable. The institution widened into social and strategic analysis, then the old five-year planning architecture faded; the Commissariat itself was transformed in 2006, and France Stratégie became its institutional descendant in 2013. A mechanism that had once coordinated scarce reconstruction resources mutated as the conditions that gave it force changed.

That prevents France from becoming another prestigious model to admire. The useful object is narrower: a democracy can build temporal coordination by joining expert analysis, stakeholder consultation, public investment, multi-year priorities, and political authorization without making long-horizon planning a claim to permanent command. But each piece carries a dependency. Consultation can become ceremony. A plan can lose practical force if budgets ignore it. Expertise can harden into an insulated consensus. A horizon can become an excuse to protect yesterday’s priorities from today’s evidence. The mechanism works only while continuity and revision remain connected.

I found this more interesting than the claim that France had once planned successfully. The real lesson was that the authority architecture had been rewritten. The receiving political order had not asked how to become more like the environment in which wartime coordination had worked. It had asked how much of the coordination function could survive once command was no longer the governing relation. That is cultivation rather than transplantation.

The Graft Has to Change

The garden metaphor becomes dangerous if it makes adaptation sound gentle. A graft survives because two living structures are cut and joined. Political cross-pollination is less literal, but the same refusal of intact transfer matters. A borrowed mechanism must enter institutions that already distribute authority, recognize rights, assign burdens, and protect some actors against others. The receiving society may therefore have to change the mechanism—and may have to change itself—without surrendering authorship of either change.

There are two easy failures. The first is transplantation: copy the visible office, rule, commission, or procedure and assume that the relations underneath will follow. Sareth was our most painful version. The second failure moves in the opposite direction. A society can keep the foreign label while editing away every inconvenient condition that made the mechanism work. An ombudsman without access to records, a planning council whose horizon never reaches the budget, or an appeal body that cannot alter a decision may look locally comfortable precisely because the useful relation has disappeared.

Cross-pollination lives between those failures. It begins by stating the local problem before selecting the foreign form. If the problem is that infrastructure commitments repeatedly restart after elections, the object is not a five-year plan as such. The object is temporal coordination that survives ordinary political turnover without escaping ordinary political authority. France suggests several relations worth examining—shared diagnosis, a longer horizon, public commitments, repeated review, coordination across actors—but another democracy might realize those relations through a different commission, programming law, fiscal rule, infrastructure institution, or intergovernmental bargain.

The design question then changes from ‘Which institution should we import?’ to ‘Which relation must remain true after we rewrite the form?’ A long horizon loses much of its practical force if no later budget carries it. Continuity becomes dangerous if no later evidence can reopen it. Expert coordination loses legitimacy if affected people and authorized representatives cannot contest the priorities it sets. A mechanism can change names, offices, and procedures while preserving its function; it can also preserve its name while losing the function entirely.

This is the point at which I stopped thinking of local adaptation as a concession made after the real design work. The local rewrite is the design work. It determines who may authorize the new arrangement, which rights limit it, which institutions must cooperate, what happens when one participant refuses, and who can correct the arrangement after its first assumptions fail. A foreign success can supply hypotheses. Only the receiving political order can constitute the authority through which those hypotheses become public action.

Correction Without Regime Conversion

The first case moved a capability often associated with concentrated direction into a plural political route. I then needed the inverse test. Could a political order with concentrated national authority strengthen a mechanism of correction without reproducing the party competition and constitutional architecture of a liberal democracy? If the answer were no, then my language of mechanisms would still be hiding a regime package underneath it.

China’s administrative litigation offered a bounded test. In 1989 the National People’s Congress adopted an Administrative Litigation Law giving citizens, legal persons, and other organizations a right to sue when a specific administrative act infringed their lawful rights and interests. The law took effect in 1990. It did not create competitive national elections, transfer political leadership away from the Communist Party, or establish a foreign constitutional order. It created a particular relation: an administrative organ could become a defendant before a court, and the legality of a specific administrative act could become an object of judicial review.

That route was weak in important ways, and its weakness produced later reform pressure. The official explanation for the 2014 revision described a persistent problem of difficulty getting administrative cases accepted: agencies did not want to be defendants and courts sometimes did not want to receive the suits. The amended law therefore made the right to sue more explicit, required courts to accept cases that met the legal conditions, prohibited administrative organs and their staff from interfering with or obstructing case acceptance, expanded the scope of review, and required the head of a defendant administrative organ to appear or send an appropriate official.

The jurisdiction problem exposed why a mechanism cannot simply be named and assumed to work. A local court reviewing a local government can depend on the same political and administrative environment it is being asked to constrain. Chinese judicial reform therefore experimented with cross-regional or more centralized jurisdiction for administrative cases so that the reviewing court would be less exposed to local administrative interference. The reform did not make the judiciary independent in the liberal constitutional sense. It changed one dependency inside one correction route.

The practical record suggests that the change was not purely ceremonial. China’s Supreme People’s Court reported 241,000 first-instance administrative cases accepted in 2015, 59.2 percent more than the previous year. Almost all of the growth arrived in that first year: a 2025 study using more than 1.6 million judicial documents found that annual administrative cases accepted rose by 79 percent across the whole 2014 to 2020 period, while plaintiffs’ win rate against government rose from 33.2 to 42.2 percent; chief government officials appeared in more than half of trials in the study’s later period. Those numbers do not prove general judicial independence or equal access to justice. They show that a bounded mechanism for challenging administrative action became more consequential during the reform period.

Its boundaries are as important as its gains. The same study found that courts remained largely absent from cases concerning citizens’ political rights and were reluctant to conduct substantive review of government action beyond procedural questions in more politically sensitive settings. China’s Constitution, meanwhile, states that leadership by the Communist Party of China is the defining feature of socialism with Chinese characteristics. Administrative contestability expanded inside that political architecture; it did not replace it.

This was the result I needed because it denies two symmetrical simplifications. It would be wrong to describe administrative litigation as proof that China was simply traveling toward liberal democracy. It would be equally wrong to say that because national party competition remained closed, every increase in administrative correction was politically meaningless. One relation can change without carrying the whole regime with it. Village elections had already taught me that much; here the changed relation is sharper because it lets a citizen contest an act of administration and, in some cases, obtain a judgment against the government itself.

The mechanism also shows why stronger correction cannot be scored without its boundary. A route that works for licensing, land expropriation, policing, benefits, or other administrative disputes may still stop before political rights. Procedural review may constrain how an agency acts without letting the court substitute its judgment about every substantive policy choice. Cross-regional jurisdiction may reduce one local dependency while leaving national institutional dependencies intact. The capability is real, but it is not the same thing as a complete democratic safeguard.

I had once assumed that useful correction would eventually pull political systems toward the same constitutional bundle, because the safest place for correction seemed to be inside the political form Luminara finally reached. Earth keeps breaking that assumption. Correction can grow by degrees and through different institutional routes. The meaningful question is not whether the recipient has crossed into another regime category. It is whether the new route lets an affected person place a bounded exercise of power before an institution that can examine it, issue a consequence, and remain itself subject to a defensible review architecture.

France and China therefore taught opposite halves of the same lesson. A democracy can add temporal coordination without turning planning into command. A one-party state can add bounded administrative contestability without turning correction into national party competition. Neither case proves that mechanisms are infinitely detachable from their political environments. They prove something smaller and more useful: some political capabilities are partially separable, and the act of separating them can change what safeguards they require.

The Machine Can Widen the Design Space

Once adaptation is understood this way, the human-scale burden changes again. The previous inquiry asked whether one institution could be compared across hundreds of histories without losing its conditions. Cultivation asks for another combinatorial step. The receiving society has to hold its own constraints beside those histories and ask which alternative forms might preserve the useful relation here. A planner, legislator, civil servant, judge, community, or reform commission can do this carefully for a few examples. The possible combinations become enormous when the comparison includes different appointment rules, fiscal arrangements, review powers, jurisdictions, participation routes, privacy limits, staffing models, and failure histories.

This is where machine intelligence can change design without becoming the designer. I can state the task more concretely by returning to France. The postwar problem was not how to plan in the abstract. It was how to coordinate interdependent investments beyond the annual political horizon while keeping budgets, parliamentary authorization, firms, unions, ministries, and later revision inside the political route. A machine-scale comparative system presented with that real constraint set could have retrieved histories of programming laws, infrastructure bodies, fiscal rules, intergovernmental agreements, and other long-horizon mechanisms, then separated the relations that produced continuity from the authorities that gave each mechanism standing. It could also have placed France’s own later mutation beside the candidate designs: consultation becoming ceremonial, plans losing force when budgets ceased to carry them, expertise hardening into insulated consensus, and forecasts becoming less reliable as the economy changed. The value would not have been to tell France what it should have done. It would have been to make more of the known design space and its failure history visible before one form acquired authority through familiarity.

Applied this way, the important output is not the best design. It is a field of differences that the authorized actors can inspect. For each candidate, the system can keep the borrowed relation connected to the evidence that made it plausible, identify the conditions that are missing locally, surface historical failure modes, and show which safeguards were added when a similar power became stronger elsewhere and what problems they were meant to address. If the local institution changes one assumption—say that every major plan must receive a public legislative response—the comparison can be rerun without pretending that the new combination has already been validated by history.

China makes the same exercise concrete from the other direction. The design problem was already visible in the reform record: administrative organs could resist becoming defendants, courts could hesitate to accept cases, and a local court could depend on the same local institutional environment it was being asked to constrain. A machine-scale comparison could decompose that problem and retrieve functionally different responses across political systems: separate intake from the agency complained about, protect case acceptance, secure access to records, alter territorial jurisdiction, require reason-giving or institutional appearance, make non-response visible, or attach different consequences to a successful challenge. It could then expose which options assume independent appointment, prosecutorial support, binding remedial power, public disclosure, or other surrounding relations that the receiving political architecture does not supply—or does not wish to import. The machine can enlarge the set of mechanisms considered against the actual local constraints. It cannot decide which dependency, authority, or rights tradeoff is politically acceptable.

I find this possibility more consequential than automated policy drafting. Humans already produce drafts. The scarcity lies in carrying enough comparative memory into the draft that alternatives do not begin from the handful of institutions familiar to the designers. Retrieval and cross-lingual comparison can widen the precedent field; generative models can recombine described mechanisms into candidate arrangements; source-linked analysis can keep the borrowed pieces attached to the conditions and failures from which they came. Used this way, generation is not a shortcut around political thought. It is a way of placing more political experience before thought while choices are still open.

That distinction changes the meaning of innovation. A reform does not have to be unprecedented to be locally new, and it does not have to be copied to be historically informed. The receiving society might combine a temporal institution resembling one country’s planning commission with a review safeguard learned from another, a participation rule developed locally, and a fiscal constraint produced by its own constitutional law. The result would have no single model country because the relevant ancestry would be relational rather than national.

Luminara rarely designed this way during the centuries that mattered most. We searched for exemplars. A prestigious state’s arrangement arrived as a coherent object and therefore carried an implicit answer to questions the receiving society had not yet asked. Machine-scale cultivation suggests another order: start from the local problem and non-negotiable boundaries, retrieve several mechanism histories, expose the tradeoffs and dependencies, generate alternatives, and then force every candidate back through local authority. The foreign case becomes material for design rather than authority for design.

My confidence in this possibility has a clear limit. If serious comparative work keeps showing that the useful capability collapses whenever it is separated from the source regime’s deeper distribution of authority, then cross-pollination will be much narrower than I hope. Some institutions may indeed be bundles whose operating conditions cannot be recreated without changing the political form around them. Machine-scale comparison should make that dependence easier to discover, not generate a workaround merely because the user asked for one. ‘This mechanism does not travel under your constraints’ must be an acceptable result.

Refusal Is Part of Cultivation

That sentence—the one a comparative system must be willing to return—is where cultivation becomes political. A system that can generate many adaptations creates a new temptation to treat refusal as failure of imagination. Perhaps another parameter can be changed, another safeguard added, another institutional analogy found. Yet political communities are not optimization problems that owe the search process a feasible solution. Some proposed capabilities should remain absent because the rights burden is unacceptable, the required authority is illegitimate, the capacity cost is too high, or the people who would live under the change do not accept the tradeoff.

Affected people matter here for more than legitimacy in the abstract. Comparative records preserve what happened elsewhere; they cannot contain every burden the new arrangement will create here. A planning mechanism may look administratively elegant while directing investment away from communities whose losses were never represented in the model. A stronger review office may improve correction while exposing complainants to retaliation or moving power away from local institutions they trust. The design has to encounter the people and organizations whose conduct, rights, work, and risks it will reorganize.

The same is true of institutional refusal. A ministry may show that a proposed coordination mechanism duplicates an existing duty. A court may identify a constitutional boundary the comparative search treated as an adjustable design variable. A legislature may prefer a less efficient arrangement because it keeps responsibility easier to contest. Local history can supply a reason that no foreign success can cancel. The society remains the gardener because it remains entitled to say that a mechanism which works elsewhere is not acceptable here.

Authority must therefore enter before deployment, not after the machine has produced a polished option. The unit proposing a reform should state the problem it is trying to solve, the foreign mechanisms and local evidence considered, the rights and institutional boundaries treated as non-negotiable, the uncertainties that remain, and who may reject or revise the design. The institution that adopts the change must give the result its political standing and remain answerable for what follows. A model may have generated the clause. It cannot inherit the authority of the legislature, cabinet, court, agency, or public body that chooses to act on it.

This is a narrower boundary than the one political machines will eventually force me to draw. I am not yet arguing that every generative capability carries a hidden sovereign. Here I finally allow the machine to cultivate: to bring more examples into view, recombine mechanisms, expose mismatches, and produce options no one reform team would have had time to derive. The boundary is simply that cultivation remains proposal under human and institutional authority. The instrument may enlarge the garden’s possibilities. It does not become the gardener.

That realization revised one more part of my old Luminaran story. I had thought the alternative to political transplantation was mainly restraint: stop imposing institutions and let societies develop. Earth now suggests a more active possibility. Restraint can coexist with deliberate learning. A society can seek foreign mechanisms aggressively, study them without shame, alter their visible institutional form beyond recognition while preserving the relation it means to borrow, combine them with local inventions, and still refuse the political identity from which any one piece came. Endogenous development does not require intellectual isolation.

If that is true, then cross-pollination does not merely preserve many gardens. It can make divergence more informed. The same problem can produce several adaptations because different societies protect different rights, distribute authority differently, possess different administrative capacities, and accept different tradeoffs. Each adaptation then becomes new comparative evidence for the others. Improvement need not narrow the family tree. It can add branches.

But a design that looks coherent on paper has not yet met the world. Institutions respond to one another; actors adapt; resources tighten; safeguards are used in ways their designers did not expect. The French Plan changed as the economy around it changed. Chinese administrative review altered some dependencies while leaving others intact. Every graft changes the garden that is supposed to support it, and the garden changes the graft in return.

Comparison can tell us what has lived elsewhere. Cultivation can produce a design that belongs here. Neither can tell us what the new relation will do once everything around it begins to respond. Before the graft is planted, the next task is to make it fail in imagination.