The Glaux

The Experiment and Its Premise

The United States was not designed as a nation in the European sense — a single people, a single culture, a single governing logic. It was designed as a federation of experiments. The framers understood, with unusual clarity, that no one knew which policies would work best under conditions of scale and diversity. The solution was not to pick the correct answer in advance but to create a system in which different states could try different approaches, observe the results, and allow the variation to teach. James Madison called this “the extended republic” — a structure large enough to contain many experiments simultaneously, and stable enough to survive the failures among them.^[1]

The underlying epistemology was Popperian before Popper: you cannot know in advance which political arrangements will produce human flourishing. You can only design a system that generates evidence and remains capable of updating on it. Competitive federalism — the idea that states compete for residents, businesses, and capital by offering different combinations of taxes, services, and regulation — was the mechanism. Citizens could vote with their feet. States that produced better outcomes would attract people and resources. States that produced worse outcomes would face pressure to reform. The system would learn.

That system is now breaking down. Not because the experiments are failing, but because the experimenters have sorted themselves into separate laboratories — and in doing so, have disabled the feedback mechanisms the design depended on.

The Sorting Machine

Between 2020 and 2024, approximately 2.5 million people left California, New York, and Illinois. Nearly as many arrived in Texas and Florida. Red states gained $37.2 billion in adjusted gross income and nearly 500,000 taxpayers; blue states lost over $40 billion and more than 500,000 taxpayers.^[2] The scale is significant. But the composition matters more than the scale.

The people leaving blue states are not a random sample of the population. They skew toward the middle class, toward small business owners, toward people for whom the cost of living has become prohibitive. The people staying in blue states skew toward the very wealthy, who can absorb the costs, and toward the highly educated, who benefit most from the dense professional networks that high-productivity cities provide. The people arriving in red states bring with them not just income but political preferences — and those preferences, in aggregate, are shifting the political character of the destination states.

This is partisan sorting: the process by which people with similar political identities increasingly cluster in the same geographic areas.^[3] It is not new — it has been accelerating since at least the 1990s — but it has reached a threshold in the 2020s where it is producing qualitatively different effects. Counties that were competitive in 2000 are now landslides. States that were swing states are now reliably partisan. The 2024 presidential election was decided in fewer than seven states out of fifty.

The sorting is self-reinforcing. When Democrats leave red states and Republicans leave blue states, the remaining populations become more homogeneous. More homogeneous populations elect more ideologically extreme legislators. More extreme legislators produce more divergent policy. More divergent policy accelerates the next wave of sorting. There is no natural equilibrium in this dynamic. It is a positive feedback loop in a system that was designed to be self-correcting.

Two Economies, One Country

The economic divergence between red and blue states is real — but it is more complicated than either side’s narrative allows.

Red states are winning the population game. Texas and Florida are growing faster than any large blue state. Their lower taxes, lighter regulation, and lower cost of living are attracting both individuals and businesses. Between 2020 and 2024, Texas added more Fortune 500 headquarters relocations than any other state. By the metric of total economic activity and population, the red-state model appears to be working.^[4]

Blue states are winning the productivity game. GDP per capita in 2025 was highest in New York ($123,369), Massachusetts ($114,635), and Washington state ($111,860) — all reliably Democratic. The industries that generate the most economic value per worker — finance, biotechnology, software, advanced manufacturing — remain concentrated in blue-state metropolitan areas. By the metric of output per person and innovation density, the blue-state model appears to be working.

Both observations are correct. They are measuring different things. The red-state model is optimised for accessibility — it produces conditions in which a larger number of people can afford to live and work. The blue-state model is optimised for intensity — it produces conditions in which a smaller number of people generate disproportionate economic value. These are not the same objective, and the tension between them reflects a genuine disagreement about what an economy is for.^[5]

The problem is not that the two models produce different outcomes. The problem is that the sorting process is preventing either side from learning from the other. A Texas that has sorted Republican will attribute its population growth to low taxes and light regulation, and will be structurally unable to see the evidence that its public education system is underfunding the next generation of workers. A California that has sorted Democratic will attribute its productivity lead to investment in education and innovation, and will be structurally unable to see the evidence that its housing policy is pricing out the middle class that sustains its social infrastructure. Both interpretations are partially correct. Neither state can update on the other’s evidence, because the people who would have updated on that evidence have already moved.

The Feedback Mechanism That Failed

Madison’s design assumed that citizens would respond to outcomes — that a state producing bad results would lose residents and revenue, face fiscal pressure, and reform. This mechanism still operates at the margin. But it has been overwhelmed by a more powerful dynamic: people are now moving not primarily in response to policy outcomes but in response to cultural and political identity.^[6]

The evidence for this is in the composition of the migration. People leaving California are not, in the main, leaving because California’s policies have failed them personally. Many of them are economically successful. They are leaving because they feel culturally alienated — because the political environment no longer reflects their values, because the public conversation has shifted in directions they find hostile, because they want to live among people who share their worldview. This is not irrational. It is a reasonable response to an environment in which political identity has become the primary axis of social sorting.

But it means that the migration is no longer generating the signal that Madison’s design depended on. When people move for cultural reasons rather than economic ones, the destination state does not learn that its policies are better. It learns that its culture is more attractive to a particular kind of person. And the origin state does not learn that its policies need reform. It learns that a particular kind of person has left. The feedback loop has been replaced by a sorting loop. The system is no longer generating evidence about which policies work. It is generating evidence about which cultures attract which people — which is a different and less useful kind of information.

The Paradox of the Homogeneous Laboratory

Here is the central paradox: the more thoroughly each state sorts itself, the more confident each state becomes in its own approach — and the less capable it becomes of evaluating that confidence.

A state legislature that is 70% one party will pass policies that reflect that party’s preferences. Some of those policies will work. Some will not. But the legislature will have difficulty distinguishing between the two, because it has eliminated the internal opposition that would have pointed out the failures. The minority party in a heavily sorted state is too weak to mount effective criticism. The media environment in a sorted state tends to reflect the dominant political culture. The civic institutions — universities, foundations, professional associations — tend to align with the majority. The feedback mechanisms that would normally correct policy errors have been weakened by the same sorting process that produced the majority.

This is the laboratory that sorted itself. The experiment is still running. But the controls have been removed.^[7]

The historical parallel is instructive. The Soviet Union conducted its own experiments in economic organisation — and for decades, those experiments appeared to be working, because the feedback mechanisms that would have revealed their failures were suppressed. The Soviet system did not collapse because it ran out of resources. It collapsed because it had disabled its capacity to learn from error. The American sorting process is not analogous in its causes or its severity. But the structural dynamic — a system that becomes progressively less capable of self-correction as it becomes more homogeneous — is recognisable.

What Could Be Tried

The instinct, when confronted with polarisation, is to call for unity — for Americans to rediscover their common ground, to lower the temperature, to find the centre. This instinct is not wrong, but it is insufficient. The sorting process is not primarily a cultural phenomenon that can be reversed by goodwill. It is a structural phenomenon produced by specific institutional arrangements, and it requires structural responses.

Ranked-choice voting and open primaries. The primary election system is the single most important driver of legislative extremism. In a closed primary with low turnout, the median voter is far more ideologically extreme than the median citizen. Ranked-choice voting and open primaries change the incentive structure: they reward candidates who can build coalitions across the political spectrum rather than those who can mobilise the base. Alaska adopted this system in 2020; the early evidence suggests it produces more moderate legislators without suppressing turnout. The constraint: the parties that benefit from the current system control the process for changing it. The failure mode: ranked-choice voting can be gamed by well-organised factions if the broader electorate remains disengaged.^[8]

Housing reform as a federalism repair mechanism. The single largest driver of blue-state out-migration is housing cost, which is itself a policy choice. Reforming restrictive zoning laws would reduce the cost-of-living pressure that drives middle-class migration, keeping a more economically diverse population in blue states and slowing the sorting process. California passed significant zoning liberalisation in 2023 and 2024; Massachusetts is following. The constraint: homeowners — who vote at higher rates than renters — benefit from housing scarcity. The failure mode: liberalisation without accompanying infrastructure investment produces density without livability.

Interstate policy learning institutions. A structured mechanism — modelled on the OECD’s peer review process — in which states systematically evaluate each other’s policy experiments and publish comparative evidence would partially restore the feedback function that sorting has degraded. The constraint: states are sovereign and cannot be compelled to participate. The failure mode: the institution becomes a venue for political point-scoring rather than genuine learning if it is not insulated from partisan control.

Deliberative polling at scale. James Fishkin’s deliberative polling experiments — in which randomly selected citizens are brought together, given balanced information, and asked to deliberate before registering their views — consistently produce more moderate and more nuanced positions than standard polling.^[9] Scaling this approach through state-level citizens’ assemblies on specific policy questions would create a mechanism for generating policy evidence insulated from the sorting dynamic. The constraint: deliberative processes are expensive and slow. The failure mode: their outputs are advisory and can be ignored by legislatures that have no incentive to act on them.

The Question the Experiment Cannot Answer

There is a version of the polarisation story that ends in dissolution — in a formal or informal separation of the country into distinct political units that no longer share enough common ground to function as a single polity. This version is not impossible, but it is probably not the most likely outcome. The United States has strong institutional, economic, and cultural integrating forces that make formal dissolution extremely unlikely. The more probable outcome is a prolonged period of functional fragmentation — in which the federal government becomes progressively less capable of acting on contested questions, in which states increasingly function as the primary unit of governance, and in which the country operates as a loose federation of increasingly distinct political cultures rather than as a unified republic.

This is not catastrophe. It may even be, in some respects, the design working as intended — a return to a more genuinely federal system in which states exercise more real autonomy. But it comes at a cost that Madison did not fully anticipate: the problems that most urgently require national coordination — climate, infrastructure, pandemic preparedness, the governance of technology — are precisely the problems that a fragmented federal system is least capable of addressing. The laboratory that sorted itself can still run experiments. It has lost the capacity to act on what the experiments reveal.^[10]

The unresolved question is not whether the sorting will continue. It will. The question is whether the institutions of American democracy can be redesigned quickly enough to restore the feedback mechanisms that the sorting process is disabling — or whether the redesign will arrive, as it usually does in American history, only after a crisis has made the cost of inaction impossible to ignore.