Legitimacy Came Before Cognition

I. The missing half of decision-making

Finland can measure GDP, employment, births, debt, inflation, and school outcomes. It can run elections, appoint ministers, audit accounts, publish statistics, and adjudicate legality. It is among the least corrupt and most legitimate states in the world.

In June 2022, the Finnish statistical authority reclassified state-subsidized housing loans, adding about €15 billion to Finland’s official EDP debt measure overnight; the classified stock has since risen to roughly €20 billion. The classification was procedurally legitimate, technically defensible under European criteria, and produced by an independent agency operating within its mandate. No public institution had accountable ownership of the consequence model — what the reclassification would do over the next decade to housing supply, fiscal stimulus capacity, and EU fiscal-framework exposure through the debt criterion — before the decision locked in.

This is the distinction the modern state hides from itself. Legitimacy tells us who may decide. Cognition tells us what the decision will do. Human civilization iterated the first for millennia and left the second fragmented.

The split is functional, not historical. Some traditions fused legitimacy and cognition through meritocratic selection or specialized oversight. Modern states fragmented cognition across statistics bureaus, audit offices, ministries, courts, academia, media, and domain regulators. Across both, the missing object is the same: public, decision-coupled, cross-domain mechanism cognition with an institutional owner.

A legitimate decision can be cognitively ownerless.

II. The anatomy of an ownerless decision

When Statistics Finland reclassified the loans funding the country’s social housing system as part of public debt in June 2022, the technical move was small and the structural consequence was large. The loans in question are repaid through tenant rents and secured by real estate at one and a half times coverage. Economically, their risk profile differs from ordinary unfunded expenditure: they are rent-repaid and secured by housing assets. Procedurally, after a Eurostat audit, they were determined to meet European criteria for state control and were consolidated into the public balance sheet. Overnight, without a single brick being laid, Finland’s EDP debt ratio rose by about six percentage points.

The downstream effects of this technical classification are political, fiscal, and architectural. Interest-subsidy loans — the state’s primary countercyclical housing tool — now increase the headline debt ratio, making them politically harder to use just as housing construction enters a deep recession. The reclassification entered the headline debt picture against which fiscal consolidation was defended. Finland’s headline EDP debt ratio now includes housing investment paid back from rent. This tightens fiscal-rule politics and debt criteria, even though the method change itself did not increase the deficit.

Markus Sovala, director general of Statistics Finland, has defended the decision and the principle behind it: statistical authorities should not weight outcomes when classifying transactions. The reasoning is correct on its own terms. The 2025 firing of US Bureau of Labor Statistics Commissioner Erika McEntarfer over weak jobs numbers is the failure mode that statistical independence exists to prevent. Sovala’s principled exclusion of consequence-weighting is exactly what his role requires.

Every actor was acting within role. That is the point.

The Netherlands, with a comparable state-controlled social-housing setup, has been contesting the equivalent Eurostat interpretation for ten years on technical grounds and continues to exclude its corresponding loans from public debt. Finland accepted the new interpretation in three. The difference is not statistical independence — both authorities are independent. The difference is that no Finnish institution owned the question of what would happen if Eurostat’s interpretation were accepted in this case. As the Yle investigation that brought the decision into public attention put it: Did no one stop to consider the consequences of the decision? The answer is unequivocal: no.

This is what the legitimacy/cognition gap looks like in 2022, in a country with high institutional trust and low corruption. It is a story about a function that nobody owned, not about bad actors.

III. Why legitimacy matured first

Why did legitimate procedure mature while mechanism cognition remained missing?

Legitimacy failures are fast, visible, and existential. A contested succession, a usurper, a constitutional rupture, a stolen election, a refused tax — these produce civil war, military coup, economic paralysis, or open rebellion within weeks to years. The cost is shared, visible, and lethal to the elites who hold power. A polity that cannot answer the question “who may decide” tends to be killed or replaced by one that can. Selection pressure on legitimacy mechanisms is intense, and it acts on the same generation that produces the failure.

Cognition failures are slow, distributed, and attribution-poor. A wrong policy theory, a poorly modeled second-order effect, an unpriced demographic constraint, an accumulated incoherence in legal architecture — these produce decay in civilizational capacity over decades. The cost is diffuse: slightly slower productivity growth, slightly worse housing supply, slightly more demographic drift, slightly more legal fog. By the time the cost becomes legible, the original decision-maker has retired, the political coalition has changed, the model has been forgotten, and the failure is absorbed by future generations rather than the present one.

Civilizations have suffered, weakened, or collapsed in part through slow cognition failures masquerading as something else: fiscal and demographic mismanagement, agricultural exhaustion, ecological overreach. It is tempting to conclude that cognition failures simply never select for institutional learning. The sharper claim survives: cognition failures rarely collapse polities fast enough for the survivors to learn the institutional lesson. By the time the cause is legible, the polity has reorganized for other reasons, and what is preserved, if anything, is broad (“don’t deplete the soil”) rather than institutional (“build a public consequence-modeling organ that would have caught the depletion before it became irreversible”).

We perfected legitimacy because bad legitimacy kills you today. We neglected cognition because bad cognition kills you tomorrow.

IV. The legitimacy stack

Across multiple civilizational lineages, governance iterated heavily on the question of who may decide. The forms differ; the function is identical.

Roman constitutional law, English parliamentary evolution, the development of constitutional courts and judicial review, electoral systems with their multiple species, separation of powers, written constitutions, succession protocols — these are the recognizable European legitimacy infrastructure built up over two millennia.

Non-European traditions carried their own legitimacy-cognition hybrids. Chinese imperial examinations selected officials through demonstrated mastery of an administrative canon. Islamic hisba and the muhtasib combined moral-legal authority with inspection of markets and public order. Mughal accounting and Ottoman defter ledgers tied revenue legitimacy to documented administration. These were real cognition functions, but they were embedded in office, domain, or moral-legal authority. The wise official could decide well; the institution did not produce a persistent, public, decision-coupled model of what the next decision would do. When the official changed, the cognition changed with them.

The cumulative effect is that across multiple traditions, the legitimacy question received centuries of institutional iteration. Polities that failed to develop legible authority lost wars, fragmented, or collapsed. The survivors, even when their substantive politics differed dramatically, share a deep institutional memory of how legitimacy is made, contested, and transferred. The legitimacy stack is mature.

V. The fragmented cognition stack

The cognition stack is not absent. It is fragmented. The pieces are mostly visible. They sort, very roughly, into four categories.

Descriptive cognition is what statistics offices, censuses, demographic registers, and government data systems produce. Statistics Finland, the US Census Bureau, Eurostat, national accounts — they are authoritative, often insulated, and structurally retrospective. They tell us what is. They do not tell us what the proposed decision will do.

Retrospective cognition is what auditors, ombudsmen, courts, and inquiries produce. Finland’s National Audit Office, the US Government Accountability Office, the UK National Audit Office, parliamentary inquiry committees — they review after the harm has occurred. By the time they report, political capital is committed, the original decision-makers have moved on, and the institutional memory required to extract a structural lesson is fading. They are vital, and they are not the missing function.

Bounded oracles are the closest things modern democracies have to decision-coupled mechanism cognition. Some, like fiscal councils — the US Congressional Budget Office, the UK Office for Budget Responsibility, the Irish Fiscal Advisory Council, the Finnish Economic Policy Council — publish models whose public credibility constrains politics. Others, like central banks, exercise delegated authority directly within a bounded mandate that already reaches across macroprudential supervision, financial stability, climate stress-testing, and systemic-risk modeling. They are politically tolerable not because their domain is intrinsically narrow but because their boundary is mandated and legitimate: monetary stability, fiscal scoring, financial resilience, or a defined safety domain.

Adversarial cognition appears where the threat is external enough to force the issue: intelligence assessment, strategic warning, military planning, and wargaming. Here the state accepts cross-domain consequence modeling because the adversary supplies a fast feedback loop. If the model fails, soldiers die, borders move, or strategic surprise imposes visible cost. This stack supports the thesis rather than refuting it. States can build cross-domain cognition when survival pressure is sharp enough. They mostly failed to generalize the capacity to domestic mechanism design, where the harms are slower, more distributed, and easier to absorb by future generations rather than the actor responsible.

The bounded oracles and the adversarial stack are evidence that decision-coupled mechanism cognition is buildable when the mandate is bounded and legitimate, or when the threat is adversarial and fast. So why did the logic never generalize? Why did civilization build mandated cognition organs for monetary policy, fiscal forecasting, and military planning, but not for welfare-state design, infrastructure maintenance, demographic sustainability, legal coherence, AI governance, or long-horizon capital stocks?

The fragmented stack delivers public facts without prediction, prediction without independence, and independent prediction without full-spectrum, cross-domain accounting. None integrates the full function. None owns the consequence-mapping function for a proposed decision before the decision locks in.

VI. Why the cognition layer stayed bounded

Why did the cognition layer never integrate, even at general scope? The technical possibility is not the binding constraint — graphing statutes, scoring proposals, and modeling consequences across domains are within easy reach of small teams using public data and ordinary engineering.

The bottleneck is institutional, not technical.

Three historical pressures and two design constraints explain why the institution never formed.

Weak selection pressure. As Section III argued, slow attribution-poor failure produces weak institutional learning. The polities that survived did not survive because they had public consequence-modeling organs; they survived for many reasons, often despite the absence. Cultures evolved heavy machinery for handling the failures that killed them quickly and light machinery for handling the failures that bled them slowly. Selection ratchets the fast loops; the slow loops drift.

Political tolerance ceiling. Cognition organs are tolerated when they are bounded, advisory, retrospective, or technically forbidding. They are resisted when they would bind cross-domain decisions before political capital is committed. The bounded oracles survive because their domain is mandated and their existence has become useful to multiple political factions across alternations of power.

Several successful conversions show a recurring pattern rather than a law. A reactive crisis or elite conflict creates demand; technical mediation gives the new function a nonpartisan form; and an available institutional vehicle lets the function attach to the state. Where conversion succeeds, one usually sees:

1. An intra-elite institutional threat that creates demand from a powerful faction for the new function.
2. Technocratic mediation that embeds independent expertise in the state apparatus and removes discretion from partisan actors.
3. A pre-existing must-pass legislative vehicle that provides the political opening.

The Congressional Budget Office (1974), the Environmental Protection Agency (1970), the FOIA amendments (1974), the Office for Budget Responsibility (2010), and the Finnish Council of Regulatory Impact Analysis (2016) each fit this pattern with varying weight. Mass public mobilization can supply pressure, but conversion usually requires an institutional vehicle and an actor with power to install the new function.

Failed conversions — post-Snowden surveillance reform, post-Cambridge Analytica platform regulation, much of post-2008 financial regulation outside the US Consumer Financial Protection Bureau — produce what Michael Power, in The Audit Society (1997), names as the default response of mature institutions to public outrage: rituals of verification. Transparency reports, oversight boards, audit trails, and compliance regimes proliferate; the underlying structural power dynamics remain intact. The illusion of accountability replaces the substance of structural reform.

Bounded and adversarial exceptions. The cognition function does survive politically when it is fenced by mandate or driven by external threat. The CBO, born from the Congress-versus-Nixon impoundment conflict, survived because it served Congress as a counterweight to the executive’s Office of Management and Budget. The US Office of Technology Assessment, defunded in 1995, is the cautionary case: advisory cross-domain organs without faction-specific indispensability are easy to cut. Project Cybersyn (Allende’s Chile, 1971–1973) is the sharper case: a cognition layer fused to a contested legitimacy project will die with that project.

The legibility constraint. A cognition layer can fail in the opposite direction: it can mistake its model for the world. Hayek’s knowledge problem and Scott’s high-modernist failure mode are not objections from outside the framework; they are design constraints on it. Mechanism cognition must not become a planner’s omniscience fantasy. It must publish assumptions, uncertainty, causal paths, excluded variables, and expected absorption points; it must import local knowledge without pretending that local knowledge has become fully centralizable; and it must compare prediction to outcome after the decision. The missing function is not a command center that decides for society. It is a public, contestable consequence model that lets legitimate decision-makers know what their proposed mechanism is claiming before they vote.

The technocracy objection. The most familiar objection to the framework is that decision-coupled cognition coupled to political action is one step from technocratic veto. If an unelected body’s models bind the political process, the legitimate decider has been displaced.

It maps the terrain; it does not choose the destination. Mechanism cognition does not decide the public goal. It states what follows from a proposed mechanism, which assumptions carry the claim, what uncertainty remains, what variables may absorb the cost, and what observation would falsify the model. Legitimacy still decides. It decides with the mechanism visible. The bounded oracles already show the principle in limited domains. Some, like fiscal councils, publish models that constrain politics through public credibility. Others, like central banks, exercise delegated authority inside a bounded mandate. In both cases, legitimacy authorizes the domain and the mandate; cognition operates inside it. What is missing is not the principle of decision-coupled cognition — the principle has been worked out and is institutionally tested — but the integrated, cross-domain, consequence-mapping function that would generalize the principle from monetary policy and fiscal forecasting to the rest of governance.

VII. Naming the function

The missing function is public, decision-coupled, cross-domain mechanism cognition with an institutional owner. It is more than expertise, debate, research, or a longer impact-assessment annex. It is the public consequence model attached to the decision before political capital locks in.

Naming the function does not build it. The historical record is clear: cognition organs survive when they fit beneath legitimacy, stay contestable, and become useful to institutions with power. The Finnish institutional proposal downstream of this diagnosis is the Mechanism Authority. That design belongs elsewhere.

This essay names the prior fact: the modern state solved authority better than consequence.

A legitimate decision can be cognitively ownerless. It no longer has to be.

Related:

• The Legitimacy Gate — the applied sibling: a signal is not yet governance until a legitimate actor must respond
• Constructive Diagnosis — the methodological standard the cognition layer is built around
• Powerless Intelligence — the AI-era variant: cognition becomes abundant before authority does
• Bad Equilibria Are Not One Thing — the failure-mode taxonomy
• The Fourth Branch — the institutional seat for cross-domain consequence cognition
• Mechanism Realism — the foundational ontology the corpus runs on