Government is an inevitable product of division of labor — not a social contract, not Hobbes’s fear, but economic specialization applied to the coordination problem.
Links: Legal Theory, Morality, Measurement, Causality, and Free Will, Opposing Forces, Economics, Computation and Information Theory, Civilizational Cycles, Risk and Entrepreneurship, Aggregation vs. Sorting
Specimens (evidence, dated):
The standard explanations for why government exists:
Chris’s argument — division of labor:
Government forms for the same reason any specialization forms: managing all interpersonal interactions personally is an unsustainable drain on human effort. Even in a purely anarchist group — no state, no rules — people must still interact constantly. Negotiating every dispute, establishing every norm, adjudicating every conflict from scratch consumes time and cognitive resources that could be spent on productive activity.
People are willing to delegate these coordination tasks to others, the same way they delegate farming, building, and medicine. The pattern scales:
| Scale | Governance Form | Delegation Mechanism |
|---|---|---|
| Extended family | Elders | Age, experience, trust |
| Tribe/village | Chief/council | Reputation, consensus |
| City-state | Monarch/assembly | Selection, tradition |
| Nation | Representative government | Election, constitutional constraints |
A monarch is just the family elder principle applied to a larger society. A representative is just a specialist to whom you’ve outsourced the coordination burden.
Pure democracy is computationally impossible. If everyone votes on everything, each citizen must be informed on every issue, evaluate every tradeoff, and participate in every decision. This is the governance equivalent of central planning — it asks every node to process all information, rather than distributing the computational load. It fails for the same structural reason central planning fails (see Computation and Information Theory): you can’t concentrate distributed knowledge at a single decision point without losing information. Representative democracy is the governance version of the price system — delegate decisions to specialists who process the relevant information full-time.
The big-L libertarian (ancap) position: Government can be replaced entirely by competing private Rights Enforcement Agencies (REAs). Since these are market-driven entities, their scope is bounded by what customers will pay for. Competition ensures quality and prevents overreach. All governance is contractual, voluntary, and subject to market discipline.
Why this is compelling: If markets outperform central planning for resource allocation (which the computation work demonstrates), why not for governance too? Competing REAs disciplined by market forces should outperform a monopoly provider. The logic tracks.
The libertarian argument for REAs: Trade is always preferable to conflict. Rational actors will always choose negotiation and arbitration over warfare because the costs of conflict exceed the costs of compromise. REAs that go to war with each other lose customers. Market incentives enforce peace.
Two REAs with conflicting judgments about a dispute between their clients have exactly three options:
This pattern has been run thousands of times in history. Greek poleis, Italian city-states, feudal lords, Hanseatic League cities — each was effectively its own REA with jurisdiction over its territory/members. The result was consistent: consolidation through either federation, conquest, or absorption by larger powers.
REAs would face the same pressures. When jurisdictions overlap — when REA-A’s client and REA-B’s client have a dispute — the system needs a resolution mechanism. Negotiation and arbitration work for low-stakes disputes. For high-stakes ones, the incentive to defect (use force) grows, especially when one party has a significant advantage.
Chris’s challenge to the libertarian peace thesis: The argument that trade always dominates conflict assumes comparable bargaining positions. But if one society has such a large force advantage over another that conquest is cheaper than ongoing trade negotiations, force is the economically rational choice.
Rome didn’t trade with Gaul — it was cheaper to conquer. The expected value of conquest (resources gained minus war costs) exceeded the expected value of trade when the military asymmetry was large enough. The libertarian counter — that conquered populations resist, maintenance costs compound, and empires eventually collapse under the overhead — is historically true as a long-run observation, but it’s an empirical claim about long-run costs, not a logical proof that trade always dominates. In the short and medium term, conquest can absolutely be cheaper. The civilizational cycle (see Civilizational Cycles) is the evidence — empires rise because force works, then fall because the long-run costs eventually overwhelm the gains.
Most goods benefit from competition. Dispute resolution is different — adjudication between two parties requires a shared authority. Two competing authorities ruling on the same dispute doesn’t produce better justice through competition; it produces deadweight loss through conflict.
The market equilibrium for dispute resolution within a territory converges toward monopoly provision. This is Weber’s definition of the state: a monopoly on the legitimate use of force within a territory. The state isn’t imposed from above — it’s the emergent equilibrium of competing jurisdiction claims.
Chris’s caveat — the monopoly isn’t “natural” in the permanent sense. No government lasts forever, which suggests the monopoly is emergent and temporary rather than “natural.” It forms because of the jurisdiction convergence logic, but it doesn’t persist indefinitely because the same power concentration that solves the jurisdiction problem creates the conditions for tyranny. The cycle: fragmentation → consolidation → stability → corruption → collapse → fragmentation. The monopoly is an attractor state, not a permanent one.
Agreement with libertarians: The unnatural monopoly on force — the fact that it trends toward tyranny — is what causes the problems. The monopoly is a practical necessity that carries inherent danger. This connects directly to the civilizational cycles work: the drift from voluntary to forced cooperation is how free societies die.
The ratchet: Organizations have asymmetric incentives around membership. Strong incentives to attract and grow — more members means more resources, more legitimacy, more power. But allowing exit introduces market discipline that constrains the organization. So organizations systematically make exit costly:
⬇ Specimen — the ratchet observed from the other side. Trolligarch’s virtual-democracy series supplies the natural experiment this section predicts: where exit is naturally free (a Discord server plus Google Docs, so revolutionaries just copy the documents and start over), no institution stabilises at all — the Republic of Kramara ran through fifteen republics, one lasting under two hours. Where infrastructure is genuinely costly (modded Minecraft servers, Habbo headquarters worth thousands of dollars), polities endured for years. He calls it the price of revolution. Read with the ratchet, the pair suggests exit cost is not merely a defect organisations impose to escape discipline — it may be a precondition for institutional durability at all, which sharpens rather than softens the ratchet’s normative sting.
The pattern is general: join easily, leave with difficulty. And once exit is foreclosed, the internal dynamics shift from “serve members to retain them” to “extract from members because they can’t leave.” This is the voluntary → coercive drift at the organizational level.
The US Civil War as the death of federalism:
The Civil War resolved the slavery question (rightly). But the structural precedent it set was: the union is permanent, states cannot secede. This transformed federalism from a voluntary association of sovereign states into a permanent union with subordinate administrative units.
The underlying contention was free association. The South no longer wanted the jurisdiction of the North. Whatever the moral dimensions of the specific dispute (and slavery is indefensible), the structural question was: can members of a voluntary union leave? The answer, established by force, was no.
Without a credible exit threat, the federal government had no structural incentive to respect state sovereignty — because states couldn’t credibly say “we’ll leave.” This is the precondition that enabled the subsequent centralization: Commerce Clause expansion, 14th Amendment incorporation, New Deal regulatory expansion, federal funding leverage. The ratchet toward federal power was unlocked when exit was foreclosed.
Hirschman’s framework — Exit, Voice, and Loyalty:
When exit is blocked, voice is the only remaining lever. But voice (voting, protest, litigation) is dramatically weaker than exit as a disciplinary force. A customer who can leave forces the company to compete on quality. A citizen who can only vote has far less leverage — especially when the feedback cycles are 2-6 years and easily captured.
This is actually the strongest argument for the REA model even though it fails on jurisdiction. If you could switch governance providers, the market discipline would prevent the degradation. The state solves the jurisdiction problem but at the cost of making exit nearly impossible — which removes the very feedback mechanism that would keep it honest. The tradeoff is structural: jurisdiction resolution vs. market discipline. The state buys the first and loses the second.
A dictatorship is a single-node architecture: maximum execution speed, zero coordination overhead. In theory, a wise and benevolent dictator is the most efficient governance structure.
It fails for three reasons:
Known failure mode: Tyranny of the majority. 51% can vote to oppress 49%. Democracy without constraints is two wolves and a sheep voting on dinner.
Why it’s still the best available option:
Moral grounding from free will. If free will is real and people are genuine causal agents (see Measurement, Causality, and Free Will), then a governance system that denies them input into their own governance is aggressing against their agency. Democracy isn’t efficient, but it’s morally required — people must have a say in their own destination.
Distributed knowledge input. Representatives aggregate knowledge from their constituents — not perfectly, but better than any single decision-maker. The legislative process is a (crude) approximation of the distributed computation that markets perform.
Error correction. Elections provide a feedback mechanism — bad leaders can be replaced without violence. This is the closest thing governance has to a market signal.
Democracy is still bad — just less bad than the alternatives. It’s slow, it’s susceptible to demagoguery, it trends toward short-term thinking (election cycles), and the majority can oppress minorities. The question is how to constrain it.
The Constitution addresses the structural problems identified above by designing a fault-tolerant distributed governance system:
Executive branch = fast execution node. Dictator-like efficiency for time-sensitive decisions (commander in chief, executive orders, treaty negotiation). But scoped — the executive executes, doesn’t legislate. A strong single decision-maker where speed matters, checked by the other branches where it doesn’t.
Legislative branch = distributed deliberation. Representative input aggregating knowledge from many nodes. The closest governance analogue to the price system — representatives “encode” their constituents’ preferences into legislation the way prices encode supply and demand. Bicameral structure (House for population, Senate for states) adds redundancy and prevents either pure majority rule or pure state equality from dominating. Note the tense: that is the designed protocol. Since the 17th Amendment (1913) the Senate is no longer a state-representing chamber — both chambers now answer to electorates, and the redundancy the design bought is largely gone (see Federalism Decay below).
Judicial branch = constraint enforcement. Ensures the protocol is followed. Resolves jurisdictional disputes — exactly the meta-arbiter function that the REA system lacks. Constitutional review means the system can check itself against its own foundational rules. This is the error-correction mechanism.
Federalism = hierarchical distribution of computation. Local decisions at the local level, escalating only when jurisdiction spans multiple localities:
This solves the REA jurisdiction problem by making jurisdiction tiered and gated — not competing horizontally, but layered vertically with defined scopes. It also preserves the distributed knowledge advantage: local governance stays close to local information, rather than routing everything through a central node.
Bill of Rights = hard constraints on the system’s output space. Certain actions are prohibited regardless of majority input. This is the structural solution to tyranny of the majority — the 51% cannot vote away the rights of the 49%. The constraints are pre-committed, not subject to democratic override (in theory — amendment is deliberately difficult).
Separation of powers = redundancy and fault tolerance. No single node failure crashes the system. Each branch checks the others. Power is distributed so that corruption in one branch is resisted by the others.
| Component | Governance Function | Distributed Systems Analogue |
|---|---|---|
| Executive | Fast execution, unified command | Single-threaded executor |
| Legislative | Distributed deliberation, knowledge aggregation | Consensus protocol |
| Judicial | Constraint enforcement, dispute resolution | Validator / consistency checker |
| Federalism | Hierarchical jurisdiction scoping | Layered network architecture |
| Bill of Rights | Hard output constraints | Protocol invariants |
| Separation of powers | Fault tolerance, error correction | Redundancy / Byzantine fault tolerance |
| Elections | Feedback mechanism, peaceful succession | Market signals / leader selection |
| Amendment process | Protocol upgrades (deliberately slow) | Consensus-required schema migration |
Government sits in the law circle of the morality framework. It is the institutional mechanism that implements codified ethics — the formal, enforceable subset. The morality/ethics/law distinction constrains what government should do:
Government is a service provider — it produces coordination, dispute resolution, and defense. Like any service provider, it’s subject to:
The Rahn Curve from the civilizational cycles work gives empirical bounds: government spending in the 15-25% of GDP range maximizes growth; above 50%, near-stagnation. The US is at ~37-39%, trending upward.
Government formation and structure is where the cycle lives. The drift from voluntary to forced cooperation is a government problem:
The constitutional experiment is the attempt to break this cycle through institutional design. Whether it works is the open question of Section X.
From the measurement-causality framework: coercion limits choice but doesn’t remove free will. Government coercion — taxation, regulation, conscription — narrows the option space available to the citizen’s simulation engine but doesn’t eliminate agency. The citizen under government still chooses, just from a constrained set.
The NAP standard: government coercion is aggression. The practical concession: some aggression (dispute resolution monopoly, defense) may be a necessary cost of coordination at scale. The constitutional project is to minimize that aggression and prevent its expansion.
The same structural pressure that would consolidate competing REAs into a state is exactly what centralizes federal systems. The mechanism is jurisdiction creep:
Historical vectors of centralization in the US:
Corollary — federalism is a capture-resistance mechanism, and the mechanism is ROI.
State-level offices are individually cheap to capture — low-information, low-turnout, small-dollar races — so the naive “decentralization raises the price of capture” claim is false, and stating it that way invites the obvious rebuttal that federalism just creates fifty bargain-bin targets. The correct variable is return, not price. Capture is an investment decision:
So federalism does not make capture hard. It makes capture unprofitable — which is the more robust protection, because it doesn’t depend on anyone’s vigilance.
The consequence worth stating plainly: cronyism and corporate capture are not primarily moral failures of particular actors. They are the predicted equilibrium once the prize becomes leveraged enough to be worth buying. That relocates the cure. Ethical appeals (better people) and regulatory patches (anti-lobbying rules — themselves capturable by the same logic) address the symptom; only de-leveraging the prize addresses the cause. This is Tullock’s rent-seeking result reached from the structural side, and it sharpens Olson’s sclerosis above from a description into a mechanism.
The Hearst case is the specimen: a media baron reaching one American in four is downstream of the Gilded Age’s national economic and informational consolidation, and the 17th extends that consolidation to the political layer — making the entire chain targetable at once, and thereby worth targeting. Connects to Press Freedom and Governance.
And this is only half of federalism’s purpose. The anti-capture surface-area argument is the defensive half; the other is proximity — governance close to the governed, so that heterogeneous polities may legitimately differ (an Alabamian can and should be governed differently from a New Yorker). That second half is the sorting case in Aggregation vs. Sorting: federalism is a partial sorting mechanism embedded inside an aggregating one. Centralization attacks both halves at once — it destroys the surface area and forces one outcome onto a heterogeneous population.
This is the city-state → nation-state consolidation pattern playing out within a federal system. The “meta-arbiter” (federal government) absorbs the jurisdiction of the lower nodes (states) through the same structural logic that makes REAs converge into a state. The Constitution was designed to resist this, but the pressures are structural, not accidental.
A constitution is source code written in natural language — inherently ambiguous. Constitutional interpretation is essentially a compiler problem: converting high-level intent into executable rules. But unlike software:
| Software | Constitutional Law | |
|---|---|---|
| Specification | Formal (types, tests, contracts) | Natural language (inherently ambiguous) |
| Bug cycle time | Days to weeks | Decades of litigation |
| Patch process | Deploy a fix | Amendment (supermajority, nearly impossible) |
| Runtime environment | Controlled, versioned | Society changes under the code |
| Bug exploitation window | Short (hotfix → deploy) | Entire duration until resolution |
| Attack surface | Can be reduced with formal methods | Every ambiguity is permanent |
“Shall not be infringed” compiles differently depending on who’s interpreting. “Regulate commerce among the several states” has been stretched from “prevent interstate trade barriers” to “regulate anything that might conceivably affect interstate commerce.” Each reinterpretation is a bug exploit — the text didn’t change, but the compiled output did.
And the interpretation process itself becomes a vector for institutional capture. Whoever controls the courts controls the compiler. This is why judicial appointments are existential political fights — they’re not about individual cases, they’re about who gets to define what the source code means.
The software developer’s frustration applies: Intent is genuinely hard to encode in language. Anyone who’s written a spec knows that edge cases emerge that the spec didn’t anticipate, and adversarial users will find every gap. Legal language has the same problem, but with century-long bug cycles and no hotfix mechanism.
Friedman’s point (applied by Sowell to education and broadly applicable): profit is a dual-signal feedback mechanism. It says both “do more of this” (profitable activities expand) AND “do less of that” (unprofitable activities contract and die). Markets self-correct because bad ideas run out of money.
Government has no equivalent loss mechanism:
This is why REAs remain appealing despite failing on jurisdiction — they’d have market feedback. A bad REA loses customers and dies. A bad government agency gets a bigger budget.
The constitutional approximation: The Founders tried to build market-like feedback into governance:
But these feedback loops are too slow and too easily captured compared to market signals. Elections happen every 2-6 years, not continuously. Federalism erodes as federal power centralizes (pressure #1). Separation of powers degrades as branches cooperate rather than compete (bipartisan spending, judicial deference to executive agencies).
The structural asymmetry: Markets adapt in real time because profit/loss is continuous and automatic. Constitutional feedback is periodic, requires active participation, and can be gamed. This speed mismatch means constitutional checks degrade faster than they can self-correct — the same structural asymmetry that makes central planning lose to markets.
Winner-take-all, single-member-district elections mathematically converge on two parties. This is Duverger’s Law, and it’s a game-theoretic inevitability, not a contingent feature of American politics. A binary outcome (win/lose) forces binary coalitions. Third parties are vote-splitters that hurt the ideologically closest major party, so strategic voters abandon them, so they die.
Multi-party systems (ranked choice, proportional representation) require a different payoff structure — multiple winning positions filled from a single election. If there are N seats and RCV fills them, suddenly third-party votes aren’t wasted. But single-seat races with a single winner will always collapse to two viable options.
The deeper point: changing the voting system doesn’t solve the governance problem. Multi-party systems (Germany, Israel, Italy) produce coalition governments that horse-trade their way to the same scope creep. More parties, same ratchet. The issue isn’t how many parties capture the state — it’s that the state is capturable. Better voting systems rearrange who holds the monopoly; they don’t dissolve it. The exit problem remains: voice, however many parties it’s distributed across, is still dramatically weaker than exit as a feedback mechanism.
Below the Duverger argument sits a harder floor: no voting system is “fair” or “optimal” in the absolute — every one is a compromise that sacrifices some reasonable property. This is the Gödel of voting:
So advocacy that promises a corner these theorems forbid — “maximizes happiness,” “no point in strategizing” (CGP Grey’s STV pitch) — is laundering a value choice as a theorem. RCV/STV is just a sequential-elimination + surplus-transfer engine; it inherits RCV’s strategy-dependence (non-monotonicity — ranking a favorite higher can make them lose, e.g. Burlington VT 2009; center-squeeze — which revives the favorite-betrayal it claims to abolish). It is a real but scope-bounded improvement on FPTP, not an escape from the floor.
Proportionality is not scale-invariant. “Make the legislature proportional” is ill-posed in a nested federation, because it’s defined relative to a reference population — district, state, or nation — and those proportionalities cannot be jointly satisfied (a delegation proportional to each state’s internal split does not compose into a nationally-proportional body). Districts are the decomposition device that makes representation tractable at a locally knowable level. The at-large alternative (e.g. California electing all its U.S. House seats in one ~100-candidate vote) fails three separate ways: mechanically (a 51% bloc can sweep every seat), epistemically (no voter can evaluate 100 candidates across the whole state), and by geographic skew (dense population centers impose representatives on rural regions they aren’t part of — the exact domination districts exist to prevent). Asking “proportional with respect to which body?” usually dissolves the pitch (scope confusion; the political instance of the accounting-identities domain-matching move).
“Fair to whom?” — proportionality is not fairness. The whole case for population-proportionality smuggles a contested value past the reader as if it were neutral. Call an all-one-species council “disproportional, therefore unfair,” and the unstated premise is population-proportionality = fairness — but ask each district and it calls its own result fair (its winner won outright); no constituent body says its result is unfair. “Unfair” is not a property of the outcome; it presupposes a reference frame the advocate never defends. Worse, the premise runs opposite to how it’s sold: pure population-proportionality amplifies the dominance of population centers (cities hold the most people), which is exactly what bounded-area representation — districts, equal-state upper houses (the US Senate’s 2-per-state, the Connecticut Compromise) — is designed to check. So an upper house being non-proportional isn’t a defect; it’s the point. Equal per-capita influence is a coherent value, but it is one value competing with anti-domination of bounded communities — and which to prefer is a normative choice, not a fact.
The upshot is normative, not technical: choosing a voting system means choosing which fairness property to sacrifice — a value judgment, not an optimization.
The blend is the answer the impossibility results force. Since no single equality can be grounded as the fair one, a mature constitution doesn’t pick one — it institutionalizes several at once and balances them across institutions. The US design is exactly this: a population-proportional House + an equal-state Senate (the Connecticut Compromise) deliberately blend the two reference frames. Presidential selection inherits the blend in its very arithmetic — the Electoral College’s 538 = 435 (House) + 100 (Senate) + 3 (DC), so each state’s weight carries both its population and a flat +2 floor. Hence the clean illustration: states pick presidents, not “the people” — and always have (in the early republic several state legislatures appointed electors directly, with no popular vote at all). The non-proportionality isn’t a flaw the founders failed to fix; it’s the compromise that makes a multi-level polity governable precisely because “fair” has no single meaning. Specimen: CGP Grey — “Too Good for Politicians to Allow” (STV).
⬆ The impossibility floor generalizes — see Aggregation vs. Sorting. That page takes the results above and adds the dimensionality variable (majority rule is well-behaved in 1 dimension via the median voter theorem, and generically has no equilibrium in ≥2 via McKelvey–Schofield), which reframes parties as a dimensionality-reduction device and the two-party collapse noted above as over-determined — Duverger’s mechanical effect plus the institutional answer to multidimensional chaos. It also names the alternative family (sorting/exit, which removes the theorems’ target rather than beating them) and the force that collapses it back (increasing returns to coordination, making law a network good — Duverger’s law and the monopolization of legal orders as one mechanism). Key correction carried there: Duverger fixes the cardinality of viable parties, not their identity, so realignment is a change of basis and a third party’s real function is displacement.