# Beyond Morality: Governance as Institutional Engineering **Niti Shakha Working Paper 0.2 — Draft for public comment. Not peer reviewed. Cite as work in progress.** --- ## Abstract Market design transformed parts of economics from a descriptive science into an engineering discipline: auction design, school matching, and kidney exchange are now built, tested, and repaired the way bridges are. This paper asks whether governance can be studied the same way. We do not claim that governance *is* an optimization problem; we investigate whether it can be fruitfully modeled as one. We propose a common vocabulary — agents, state, information, objectives, mechanisms, institutions, feedback — precise enough that claims about governance become checkable. Using this vocabulary we develop two propositions. First, norm-based governance of the kind documented by Ostrom depends on information conditions (cheap mutual monitoring, repeated interaction, reputation-carrying communication) that degrade systematically with scale. Second, when those conditions degrade, electoral selection begins to resemble a market with asymmetric information about a latent quality — governance competence — and is therefore exposed to adverse selection: a *political lemons* problem. Both propositions are stated as conditional, falsifiable claims, and we identify the existing evidence that constrains them, including evidence that cuts against them. We close with a list of open research problems. The purpose of this paper is not to settle any of these questions but to define a research program in which they can be settled. **Keywords:** institutional design, mechanism design, market design, commons governance, adverse selection, political selection, scaling, distributed optimization **Status:** Working Paper 0.2. This document exists to stabilize concepts, invite criticism, and coordinate further work. Sections will change. If you find an error in logic, a missing literature, or a sharper counterexample, that is exactly the feedback this draft is for. --- ## 1. Introduction: The Economist as Engineer In 2002 Alvin Roth observed that economics had quietly acquired a second job. Beyond describing how markets behave, economists were increasingly asked to *build* them: to design spectrum auctions, match medical residents to hospitals, and construct exchanges through which incompatible kidney donor–patient pairs could trade. Roth called this work "the economist as engineer" (Roth, 2002). The distinguishing feature of engineering, he argued, is that it must deal with the messy details that theory abstracts away — and that doing so generates new theory rather than merely applying old theory. Market design succeeded because it combined three things: a formal apparatus (game theory and mechanism design), a willingness to test designs before deployment (laboratory and field experiments, computation), and an institutional pathway from design to adoption. The results are concrete. The National Resident Matching Program processes tens of thousands of physicians annually through an algorithm whose incentive properties are mathematically characterized. Kidney exchange chains, impossible under the prior institutional arrangement, now save lives routinely (Roth, Sönmez & Ünver, 2004). This paper asks a question that we believe is natural but is rarely posed in this form: > **Can governance itself become an engineering discipline?** Note the phrasing. We do not assert that governance *is* an optimization problem, a control system, or a market. Assertions of that kind have a bad history: they tend to flatten what they model and to license overconfident intervention (Scott, 1998). We ask instead whether governance can be *fruitfully modeled* as an engineering problem — whether the modeling choice yields explanations, predictions, and design guidance that existing framings do not. Three observations motivate the question. **First, virtue-based reform has a poor track record as an engineering strategy.** Political discourse across democracies returns persistently to the language of character: honest leaders, selfless servants. Yet the mechanisms of politics select for the ability to win elections — fundraising, coalition management, media presence — which need not correlate with the ability to govern. Actors who arrive virtuous face systematic incentives to behave otherwise, and personality-dependent movements collapse with the personality. None of this proves virtue irrelevant; it suggests that virtue is an input the system cannot reliably procure, and that a design which requires it is a design with an unsourced dependency. **Second, the failures are patterned, which suggests structure.** Governance failures repeat across countries, cultures, and centuries with a regularity that character-based explanations cannot absorb. Commons are overgrazed, principal–agent slack accumulates, coordination fails at predictable junctures. In India — the setting that motivates our own work — a child born to a father in the bottom half of the education distribution can expect to reach only around the 38th percentile of the national education distribution, compared with roughly 42 in the United States and 47 in Denmark, and this figure has been largely flat for decades (Asher, Novosad & Rafkin, 2024). A constant, cross-generational outcome under rotating leadership is evidence about the *system*, not about the individuals passing through it. **Third, the intellectual raw material already exists but is scattered.** Mechanism design tells us how to align private incentives with designed outcomes. Institutional economics tells us how rules shape long-run performance. Public choice tells us how political actors respond to incentives. Control theory tells us how feedback stabilizes or destabilizes systems. Complex-systems research tells us how local interactions produce global patterns. Each of these disciplines holds a piece of the governance problem. What is missing is not insight but a *common abstraction* within which the pieces compose. The remainder of the paper proceeds as follows. Section 2 surveys the existing paradigms and states precisely what we believe they lack. Section 3 proposes a minimal formal vocabulary. Section 4 introduces the first organizing distinction of our program: the separation of objective selection from institution design. Section 5 defines scale and the information quantities that vary with it. Section 6 reinterprets Ostrom's design principles as solutions that are optimal under specific information conditions, and asks what happens when those conditions fail. Section 7 develops the political lemons proposition. Section 8 sketches the integrating abstraction — governance as a distributed optimization system — that we believe can host all of the above. Section 9 lists open research problems. Section 10 concludes with an invitation rather than a claim. Throughout, we hold ourselves to a discipline borrowed from mathematics. Every section should answer, implicitly or explicitly: What exactly is the object under study? Why is the existing abstraction insufficient? What new abstraction is proposed? What explanatory power does it gain? And how could someone prove it wrong? --- ## 2. Existing Paradigms We briefly characterize the traditions that study governance, in each case identifying what the tradition explains well and where its abstraction stops. The survey is deliberately compressed; the purpose is not review but triangulation. **Political philosophy** asks what governance is *for*: what justice requires (Rawls, 1971), how competing conceptions of the good should be weighed (Sen, 2009), what legitimates coercive authority. Its strength is normative depth. Its abstraction stops at implementation: a theory of justice is not a theory of how institutions built by self-interested agents will actually behave. **Institutional economics** treats institutions — "the rules of the game" — as the primary determinant of long-run economic performance (North, 1990). It explains why societies with similar endowments diverge, and why inefficient institutions persist. Its abstraction is powerful but largely retrospective: it identifies which rules mattered after the fact, with limited machinery for *deriving* rules that will work in advance. **Mechanism design** is the inverse of game theory: given a desired outcome, construct the game whose equilibrium delivers it (Hurwicz, 1973; Myerson, 2008; Maskin, 2008). It provides the deepest available results on what alignment between individual incentives and collective goals is achievable — and, through impossibility theorems, what is not. Its abstraction stops at the boundary of the mechanism: it typically assumes the objective is given, the designer is benevolent, participation is well-defined, and the mechanism itself will be faithfully executed. Governance violates every one of these assumptions somewhere. **Public choice** applies economic self-interest symmetrically to political actors: voters, legislators, bureaucrats (Buchanan & Tullock, 1962; Olson, 1965). It predicts rent-seeking, capture, and the collective-action failures of large groups. Its abstraction is corrosive in the useful sense — it removes the assumption that officials are benevolent — but it is stronger as critique than as construction. **Computational social science and agent-based modeling** simulate societies of interacting agents to study emergent institutional phenomena. They can represent heterogeneity and disequilibrium dynamics that closed-form theory cannot. Their abstraction stops at validation: it is hard to know when a simulation's failure modes are the world's. **Control theory** studies systems steered toward objectives under feedback, with formal treatment of stability, delay, observability, and robustness (Åström & Murray, 2008). Almost none of its machinery has been seriously applied to political institutions, despite the fact that "steer a system toward a target under noisy, delayed feedback" is a fair description of what governments attempt. **Complex-systems research** studies how local interaction rules generate global structure — power laws, cascades, phase transitions. It warns, correctly, that interventions in coupled systems have non-local effects. Its abstraction stops at design: emergence is described, rarely engineered. Each tradition, we suggest, is looking at a different projection of the same object. And this motivates the claim on which the rest of the paper rests: > **These disciplines describe complementary aspects of governance, but they lack a common abstraction.** There is no shared formal vocabulary in which a Rawlsian objective, an Ostrom monitoring rule, a mechanism-design incentive constraint, and a control-theoretic stability condition can be stated as claims about the same system. Constructing that vocabulary is the first task of the research program. We begin it now. --- ## 3. Governance as Institutional Engineering: Definitions This section fixes terminology. The definitions are informal in presentation but are intended to be formalizable; each names a distinct object, and later sections use them load-bearingly. We define nothing we do not use. **Definition 1 (Agent).** An *agent* is any entity that observes, decides, and acts within the system: an individual, a firm, a party, a bureau, a court. Agents have private information and preferences that need not align with any collective goal. **Definition 2 (State).** The *state* of a governed system at a time is the vector of conditions the system cares about or that determine its dynamics: resource stocks, income and education distributions, infrastructure condition, trust levels, the current assignment of offices to agents. The state is never fully observable by any single agent. **Definition 3 (Information structure).** The *information structure* specifies, for each agent, what that agent can observe about the state and about other agents' past actions, at what cost, with what noise, and with what delay. Two systems with identical rules but different information structures are different systems. **Definition 4 (Objective).** An *objective* is an ordering over states (or trajectories of states): a specification of which conditions of the system count as better. Objectives may be contested, vague, or plural; Section 4 treats their selection as a problem in its own right, distinct from the design of institutions that pursue them. **Definition 5 (Policy).** A *policy* is a mapping from observed information to actions, executed by some agent or body holding authority: given what is seen, what is done. **Definition 6 (Mechanism).** A *mechanism* is a formally specified procedure that takes messages or actions from agents and produces an outcome: an auction, a voting rule, an examination, a matching algorithm. Mechanisms are the precise, analyzable components of institutions. **Definition 7 (Institution).** An *institution* is a persistent structure of rules, roles, and enforcement that constrains and channels agent behavior over time (North, 1990). An institution typically bundles several mechanisms with informal norms, staffing processes, and enforcement capacity. Institutions are to mechanisms roughly what deployed systems are to algorithms. **Definition 8 (Feedback).** *Feedback* is the channel by which information about realized states returns to the agents or bodies whose policies produced them — elections, audits, price signals, protests, published statistics — together with that channel's noise, delay, and manipulability. **Definition 9 (Governance).** *Governance* is the joint operation of institutions, policies, and feedback by which a population of agents steers the state of its shared system with respect to some (possibly contested) objective. Two remarks. First, nothing in these definitions presumes democracy, markets, or any particular institutional form; a village commons, a national election, and a regulatory agency are all instances. Second, the definitions expose immediately where governance differs from textbook mechanism design: in governance the objective is contested (Definition 4), the information structure is scale-dependent (Definition 3 — the subject of Section 5), and the feedback loop routes through the same agents the mechanism is meant to constrain (Definition 8). These three deviations, we will argue, are not nuisances; they are the problem. --- ## 4. The Objective–Mechanism Separation The first organizing claim of our program is a separation: > **The selection of objectives and the design of institutions are distinct problems, requiring distinct methods, and conflating them is a standing source of confusion in governance debates.** *Objective selection* is the problem of Definition 4: deciding which orderings over social states the system should pursue. This is the home terrain of political philosophy and social choice. Rawls (1971) proposes a procedure (choice behind a veil of ignorance) that outputs objectives (liberty, then the difference principle). Sen (2009) argues that complete orderings are unnecessary — comparative judgments of manifest injustice suffice — which in our vocabulary is a claim about how much objective specification institution design actually requires. Arrow (1951) proves that no aggregation rule converts arbitrary individual orderings into a collective ordering while satisfying a short list of reasonable axioms — which in our vocabulary is a *constraint on objective selection*, often misread as an impossibility result about democracy itself. *Institution design* is the problem of Definitions 5–8: given an objective, construct mechanisms, institutions, and feedback such that self-interested agents operating under the actual information structure move the state toward it. This is the home terrain of mechanism design and market design. The separation matters for three reasons. **It localizes disagreement.** Much governance debate is unproductive because parties disagree simultaneously about ends and means without noticing which. The separation allows a discourse in which one can say: *we disagree about the objective; conditional on your objective, here is what the design evidence says.* This is precisely the move that made market design tractable — the designer of a school-matching mechanism does not adjudicate what a good school is; she takes stated preferences and optimizes subject to them. **It reveals a division of labor that already implicitly exists.** Constitutions predominantly encode objectives and constraints (rights, directive principles, federal structure); legislation and administration predominantly encode policies and mechanisms. Treating a constitution as an *objective specification* — a formal statement of what the system is for, against which institutional performance can be evaluated — is, we believe, a productive reframing, and appears in Section 9 as an open problem. **It exposes an asymmetry in maturity.** Objective selection has been debated for twenty-five centuries with great sophistication. Institution design as a *forward* engineering discipline — derive the institution from the objective and the information structure, predict its failure modes, test before deployment — barely exists outside auction houses and matching clearinghouses. The asymmetry suggests where the marginal research effort should go. One caution. The separation is analytic, not absolute: institutions shape the preferences and beliefs of the agents within them, so long-run objective selection is partly endogenous to institutional design. We flag this now and return to it in Section 9; it complicates the separation but does not, we think, dissolve its usefulness, any more than plant–controller separation is dissolved in control engineering by the existence of plant–controller interaction. --- ## 5. The Scaling Problem Governance arrangements that work at one population size routinely fail at another. This observation is ancient; our aim is to state *why* precisely enough to make it a theory rather than a proverb. We first define the quantities that vary with scale. Only then (Section 6) do we bring in the empirical literature. **Definition 10 (Scale).** The *scale* of a governed system is the number of agents N whose behavior the system's institutions must constrain, together with the geographic and social dispersion of those agents. **Definition 11 (Observability).** The *observability* of an agent's behavior is the probability that a deviation from a rule by that agent is detected by some other agent who has both the incentive and the standing to respond, per unit of monitoring cost expended. **Definition 12 (Monitoring cost).** The *monitoring cost* of an institutional arrangement is the total expenditure — attention, time, infrastructure — required to sustain a given level of observability across all agents. **Definition 13 (Communication cost).** The *communication cost* is the expenditure required for information held by one agent (in particular, reputational information: who deviated, who can be trusted) to reach the agents whose decisions it should inform, with acceptable fidelity and delay. **Definition 14 (Coordination cost).** The *coordination cost* is the expenditure required for a group of agents to agree on a joint action — including agreement on whether a violation occurred and what sanction it merits. With these definitions, the scaling problem can be stated as a set of claims about how these quantities move with N. **Claim 5.1 (Monitoring does not scale).** In a community where agents interact repeatedly and observe one another in the course of ordinary life, observability is a *byproduct* — its marginal cost is near zero. As N grows and interactions become anonymous and non-repeated, observability must be *produced* deliberately, and the monitoring cost of holding it constant grows at least linearly in the number of relationships that matter, which grows faster than N. **Claim 5.2 (Reputation does not scale).** Human agents can track detailed social information — who did what to whom, who is reliable — for a bounded number of others; converging anthropological and neurological evidence puts the bound in the low hundreds (Dunbar, 1993). Below the bound, reputational information propagates through gossip at near-zero communication cost and deters deviation because future interaction is likely. Above it, most agent pairs will never interact again, reputational information decays before it arrives, and the deterrent force of "everyone will know" collapses. **Claim 5.3 (Sanctioning does not scale).** Informal sanctions — disapproval, exclusion, loss of standing — are cheap to administer and severe in consequence inside a dense social network, because the sanctioned agent cannot exit the network at low cost. At scale, exit is cheap, the sanctioners are strangers whose disapproval carries no weight, and coordination cost (Definition 14) makes even agreeing on the facts expensive. Olson (1965) established the corresponding group-size logic for contribution to collective goods: in large groups, each agent's share of the benefit from her own compliance tends to zero while her cost of compliance does not. **Claim 5.4 (Internalized norms partially scale, but their enforcement does not).** Guilt, shame before an imagined audience, and internalized duty operate without external monitoring and therefore survive scaling better than Claims 5.1–5.3 alone would predict. But internalized norms are themselves maintained by an ambient enforcement environment — they decay across generations when visibly unenforced, because agents observe that deviation is common and profitable. Internalization buys time; it does not repeal the claims above. The joint import of Claims 5.1–5.4 is a statement we will rely on repeatedly: > **Moral and norm-based governance is not wrong; it is an engineering solution whose operating envelope is bounded by the information structure of small groups.** Outside that envelope — beyond the scale at which monitoring is a byproduct, reputation propagates, and sanctions bind — the solution does not merely weaken. Its *load-bearing assumptions* are absent, and something else must carry the load. What has historically carried the load is *formal institutions*: law, courts, records, professional enforcement. The economic-history literature documents the transition directly. Greif (1993) shows how the Maghribi traders sustained long-distance honesty through a closed reputation coalition — and how that solution capped the scale of their trade. Milgrom, North & Weingast (1990) show how the medieval law merchant substituted an institution — a court that recorded judgments and made reputational information queryable — for the community's memory, exactly the replacement of Claim 5.2's gossip channel by designed infrastructure. The pattern is general: **when scale destroys an information condition, successful societies have built an institution that reproduces the condition's function artificially.** The scaling problem, so stated, is a falsifiable framework: it predicts *which* governance arrangements fail at scale (those whose enforcement relies on byproduct observability and reputational propagation) and *what* successful replacements have in common (they manufacture observability, memory, or sanctioning capacity institutionally). Section 6 tests this framing against the richest available body of evidence on small-scale governance. Section 7 applies it to the institution modern societies use to select their governors. --- ## 6. Ostrom Through an Information Lens Elinor Ostrom's *Governing the Commons* (1990) demolished a simple dichotomy. Communities around the world — Swiss alpine pastures, Japanese village forests, Spanish irrigation systems, Filipino zanjeras — had governed common-pool resources for centuries without either privatization or central state control, contradicting the fatalism of Hardin (1968). From extensive fieldwork, Ostrom distilled eight design principles associated with durable self-governance, including: clearly defined boundaries; congruence between rules and local conditions; participation of resource users in modifying rules; monitoring by accountable monitors; graduated sanctions; and cheap, local conflict-resolution forums (Ostrom, 1990; 2010). We propose a reinterpretation. We do not say Ostrom was wrong; the fieldwork is unimpeachable and the principles are among the best-validated findings in institutional analysis. We say instead: > **Ostrom's design principles can be read as a specification of the information conditions under which norm-based governance is feasible — and each principle maps to one of the quantities defined in Section 5.** Consider the mapping. - *Clearly defined boundaries* fix the set of agents, making reputational bookkeeping (Claim 5.2) tractable: you cannot track the standing of an unbounded population. - *Monitoring by accountable monitors* — in Ostrom's cases, typically the resource users themselves, monitoring as a byproduct of use — is Claim 5.1's near-zero-marginal-cost observability, achieved because appropriators are physically present at the resource. - *Graduated sanctions* exploit the fact that in a repeated game among identified agents, a mild sanction carries information ("you were seen") whose deterrent value exceeds its material sting; this requires exactly the dense-network conditions of Claim 5.3. - *Cheap local conflict resolution* is Definition 14's coordination cost held down by proximity and shared context. - *Congruence with local conditions* and *participation in rule modification* keep the feedback loop (Definition 8) short: the people who observe the state of the resource are the people who adjust the rules. On this reading, Ostrom's principles are not a refutation of the scaling problem but its most precise confirmation: they enumerate what must be true of the information structure for governance-without-Leviathan to work. Ostrom herself was explicit that her cases were small — typically 50 to 15,000 appropriators — and she resisted the "panacea" reading of her own work, insisting that institutional solutions are contingent on context (Ostrom, 2010). The productive question is therefore not whether Ostrom's principles are correct — they are — but: > **What happens when the assumptions no longer hold?** What replaces monitoring-as-byproduct when the monitored activity is a ministry's procurement rather than a neighbor's grazing? What replaces gossip when the relevant public is four hundred million voters? What replaces graduated sanctions when the deviator can exit into anonymity? Three families of answers appear in history and in the literature, and they organize much of the remaining research program: 1. **Nesting.** Ostrom's eighth principle — nested enterprises, layered governance for larger systems — points toward federalism: keep each layer's problem small enough that local information conditions hold, and design the interfaces between layers. What is preserved and what is lost in the interfaces is, in our view, an open formal question (Section 9). 2. **Institutional manufacture of information.** The law-merchant pattern (Milgrom, North & Weingast, 1990): build registries, courts, audits, published records — infrastructure that performs artificially the functions that observation and gossip performed naturally. Modern analogues include credit bureaus, disclosure regimes, and public statistical systems. The design question is which information functions must be manufactured, at what fidelity, and how the manufacturing institution is itself kept honest. 3. **Selection interfaces.** Where the governed system must delegate — where a few agents will act for many — the information burden concentrates at the moment of *selection*: choosing which agents receive authority. Elections are the canonical selection interface of modern governance. They are also, we argue next, the point at which the scaling problem bites hardest. --- ## 7. Political Lemons: Adverse Selection in Electoral Delegation Akerlof (1970) analyzed markets in which quality is known to sellers but not to buyers. His result: when buyers cannot distinguish good used cars from bad, they rationally offer a price reflecting average expected quality; owners of good cars withdraw; average quality falls; the price falls further; in the limit, only lemons trade. The mechanism requires three ingredients — a latent quality, an information asymmetry about it, and a transaction whose terms cannot condition on the truth. We now develop, carefully, the proposition that large-scale electoral delegation contains the same three ingredients. We proceed step by step because the argument is an *extension* of Akerlof, not an application, and the differences matter. **Step 1: Governance competence is a latent quality.** The capacity of a candidate to govern well — to analyze policy, administer institutions, resist capture, weigh long-run consequences — is real, variable across candidates, and known far better to the candidate (and her close associates) than to any voter. This is Definition 3's information asymmetry applied to the selection interface. **Step 2: Voters observe signals, not quality.** What reaches the voter is a heavily processed signal: speeches, advertisements, rally performances, media coverage, party labels, endorsements. Each of these is produced strategically, mostly by agents whose payoff depends on the *impression* created rather than the truth conveyed. **Step 3: Campaigns, parties, and media are signaling institutions.** In Spence's (1973) terms, a signal separates high types from low types only if it is differentially costly — cheaper for the genuinely competent to send than to fake. The critical question for each political signaling channel is therefore: *is its cost structure quality-correlated?* Charisma, fundraising ability, and media savvy are costly signals of *something*, but the something is campaigning ability, not governing ability. A party label can be a genuine quality signal — if the party screens internally. This observation will matter below. **Step 4: Scale degrades signal quality.** Here the argument connects to Section 5. In a small polity, voters have direct or one-hop knowledge of candidates: the signal channel is the dense reputational network of Claim 5.2, which is expensive to fake at the source. As the electorate grows, direct knowledge vanishes; the signal must travel through mass media and advertising; and the channel's capacity is consumed by whoever spends most on it. The information available per voter about latent quality *falls* with scale, while the cost of fabricating a favorable signal is roughly scale-invariant (one advertisement serves millions). The signal-to-noise ratio of the selection interface is therefore a decreasing function of N. **Step 5: The transaction cannot condition on the truth.** A vote is exchanged for promises. Unlike a used-car sale, there is no warranty: no enforceable mechanism by which the elected agent's performance is settled against the representations made. Retrospective voting — punish incumbents for bad outcomes — is the closest analogue, but it operates on outcomes that are noisy, delayed, and attributable only with difficulty (Ashworth, 2012). **Proposition (Political Lemons, conditional form).** *In an electoral system where (i) governance competence is latent, (ii) the dominant signaling channels have cost structures uncorrelated with competence, and (iii) no institution performs pre-electoral screening of quality, the expected competence of the candidate pool declines as electorate scale increases — because the return to investing in genuine competence falls relative to the return to investing in signal production, and agents with high opportunity costs (the competent) increasingly select out of candidacy.* Three features of this proposition deserve emphasis. **It is conditional, and the conditions are the interesting part.** The political selection literature contains results that appear, at first reading, to refute any lemons story. Dal Bó, Finan, Folke, Persson & Rickne (2017) show, with exceptional Swedish administrative data, that Swedish politicians are *positively* selected — scoring higher on cognitive and leadership measures than the populations they represent, without being socioeconomically unrepresentative. We regard this not as a refutation but as the strongest available evidence about the *conditions*: Sweden combines strong parties that screen candidates internally (restoring condition iii), a high-information media environment, and low returns to office relative to outside options for the talented. Where the literature examines weaker screening environments, the picture darkens: Caselli & Morelli (2004) model low-quality citizens' comparative advantage in seeking office when rents are high and voters ill-informed; Besley (2005) frames political selection as a neglected first-order problem. In India, self-sworn affidavits show 93% of Lok Sabha winners in the 2024 general election held family assets exceeding one crore rupees — consistent with a selection interface at which resources for signal production, rather than latent competence, are the binding requirement for entry. The proposition thus predicts *cross-system variation*: adverse selection intensity should covary with electorate size per screening institution, media independence, and the rent value of office. That prediction is testable, and Section 9 lists it. **It differs from Akerlof structurally, not just in subject matter.** In the used-car market, the good types *exit* and the market thins. In elections, the "market" cannot thin — some candidate always wins. Adverse selection in governance therefore manifests not as market collapse but as *pool degradation with full throughput*: the system continues to operate at capacity while the quality distribution of what it selects deteriorates. This is arguably worse than the Akerlof case, because the standard signal of market failure (volume collapse) never appears; the failure is silent. Second, the buyers (voters) are also the parties harmed by everyone else's purchase — a vote is a purchase with externalities — so individual voters lack incentive to invest in discrimination even when discrimination is possible (rational ignorance; Downs, 1957). Both differences make the electoral version *more* prone to the pathology than the market version, not less. **It identifies the design surface.** If the proposition is right, the remedies live exactly where the conditions live: institutions that restore screening (condition iii — professional licensure is the analogue in medicine and law), channels whose cost structure correlates with competence (condition ii — structured public examination of reasoning, track-record disclosure with verification), and feedback that conditions on measured outcomes rather than on impressions (Section 8). We do not develop remedies in this paper; we note only that the proposition converts "politics attracts the wrong people," a complaint as old as Plato, into a claim about identifiable, manipulable conditions. --- ## 8. Toward Governance as a Distributed Optimization System Sections 5–7 analyzed two components — norm enforcement and electoral selection — with the vocabulary of Section 3. This section states the abstraction that we believe unifies the components, and that gives the research program its shape. > **Modeling proposal.** A governed society can be modeled as a *distributed optimization system*: a population of self-interested agents (Definition 1), embedded in an information structure (Definition 3), whose interactions — structured by institutions (Definition 7) — collectively steer a partially observable state (Definition 2) with respect to a contested objective (Definition 4), under feedback (Definition 8) that is noisy, delayed, and endogenous. Under this proposal, the traditions of Section 2 become descriptions of *components* of one system: - **Objectives** (political philosophy, social choice) specify the loss function — including the second-order problem that the loss function is aggregated from agents who disagree (Arrow, 1951). - **Incentives** (mechanism design, public choice) determine whether each agent's local optimization is aligned with or opposed to the global objective; incentive compatibility is the condition under which the distributed system does not fight itself. - **Information** (Sections 5–7) determines what any component *can* compute: observability bounds what can be enforced, and signal quality at selection interfaces bounds the competence the system can recruit. - **Institutions** are the system's architecture: they allocate authority (which agent computes which policy), manufacture information (registries, audits, statistics), and encode the constraints that keep local optimizers within envelope. - **Computation** is a genuine constraint, not a metaphor: finding equilibria, aggregating preferences, and evaluating policies are computationally hard problems in the formal sense (Nisan, Roughgarden, Tardos & Vazirani, 2007), and Hayek's (1945) argument against central planning is precisely a claim that the required computation is infeasible without distributing it through a price mechanism. - **Feedback** closes the loop — and control theory tells us that loops with long delays and noisy sensors oscillate or destabilize unless the controller is designed for those delays (Åström & Murray, 2008). Elections are a feedback channel with a multi-year sampling period, one-bit-per-voter resolution, and a sensor (public impression) that the controlled entity can partially rewrite. Stated this way, the chronic short-termism and cyclical overcorrection of democratic policy are not moral failures; they are the expected dynamics of a poorly compensated control loop. - **Emergence** (complex systems) is the warning label: the plant is adaptive, the agents learn, and any deployed mechanism becomes part of the environment agents optimize against — the general form of Goodhart's law. A worked informal example makes the frame concrete. Consider a state government charged with improving educational opportunity. The *objective specification* lives in constitutional directives and statutes. The *sensors* are administrative data systems, surveys, and a census — each with known noise and manipulation surface. The *controller* is a legislature and bureaucracy whose members are themselves agents with careers, i.e., the controller has its own objective function, which is the deepest disanalogy with engineered control systems and the reason mechanism design, not control theory alone, must sit at the center of the discipline. The *actuators* are budgets, staffing, and law. The *feedback delay* is generational: an intervention in early-childhood education registers in earnings data twenty years later, several electoral cycles after the deciding coalition has left office. No engineer would expect a system with this loop structure to converge without explicit design for delay — intermediate observables, pre-registered causal models, and institutional memory that outlives the controller's personnel. Whether such design is possible for real polities is, in our view, the central question of the program. We stress what the proposal does *not* claim. It does not claim that societies have a single objective, that agents are rational optimizers in the narrow sense, or that governance reduces to an algorithm. It claims that the *concepts* of distributed optimization — local objectives versus global objectives, information constraints, incentive alignment, feedback stability, computational feasibility — carve governance at joints that existing vocabularies do not, and that theorems and measurements become possible once the carving is made. The claim is methodological, and it will be judged by output: either the frame produces propositions that survive testing (Section 9) or it joins the long list of governance metaphors. --- ## 9. Open Research Problems A research program is defined by its open problems more than by its claims. We list the ones we consider most important, phrased so that each could anchor a self-contained piece of work. We invite collaboration — and refutation — on all of them. **P1. Formalizing the scaling limit.** State Claims 5.1–5.4 in a model: agents on a social network, monitoring as a byproduct of interaction, reputation as information diffusion, sanctions as network exclusion. Derive the scale threshold at which cooperation sustained by these forces collapses, as a function of network parameters. Ostrom's design principles should emerge as the optimal configuration below the threshold; the law-merchant pattern should emerge as the constrained optimum above it. (Sections 5–6.) **P2. Testing political lemons.** The conditional proposition of Section 7 predicts that measured politician quality falls with electorate size per screening institution, media independence, and office rents. Candidate-level data exist: Swedish registries (Dal Bó et al., 2017), Indian affidavit data on assets, criminal cases, and education. Does within-country variation in constituency size and party screening predict quality gradients in the direction the proposition requires? What observable proxies for latent competence are defensible? (Section 7.) **P3. Observability metrics.** Define and measure the *observability* (Definition 11) of real governance institutions: what fraction of consequential official actions are visible, to whom, at what delay, with what audit trail? A rigorous observability metric would do for institutional transparency what error bars did for measurement — convert a slogan into a quantity. (Sections 3, 5.) **P4. Constitutions as objective specifications.** Can a constitution be usefully formalized as an objective specification (Section 4) — a set of orderings and constraints against which institutional performance is evaluable? What is lost in the formalization? The exercise would expose where existing constitutions specify objectives precisely, where they delegate objective selection, and where they are silent. (Section 4.) **P5. Control-theoretic analysis of electoral feedback.** Model the election cycle as a sampled-data control loop: multi-year sampling, low-resolution measurement, sensor manipulation by the plant. Characterize the stability and steady-state error of policy under this loop, and the value (in objective terms) of adding intermediate sensors — audited statistics, mandatory outcome disclosure — between elections. (Section 8.) **P6. Reproducing institutional failure in silico.** Build agent-based models in which known failure modes — commons collapse beyond the Ostrom envelope, adverse selection in delegation, capture of monitoring institutions — emerge from the information and incentive structures rather than being assumed. A simulation that reproduces the *conditions* under which Sweden selects well and weaker screening environments select badly (Section 7) would be substantial evidence that the frame captures the mechanism. (Sections 6–8.) **P7. Cooperation stability across generations.** Claim 5.4 asserts that internalized norms decay when visibly unenforced. What is the decay rate? Under what institutional conditions do internalized norms and formal enforcement complement rather than crowd out each other? The crowding-out literature in behavioral economics is the starting point; the governance-scale version is unwritten. **P8. The interface problem in nested governance.** Ostrom's nesting principle keeps each layer small, but the interfaces between layers (local–state–national) carry aggregated information and delegated authority. What do the interfaces necessarily lose? Is there a formal sense in which some objectives are *unimplementable* through any nesting — an impossibility result for federalism? (Section 6.) A top journal might regard this section as an admission of incompleteness. In a working paper, it is the opposite: it is the program. --- ## 10. Conclusion We have proposed that governance be studied as institutional engineering: a discipline whose object is the design of institutions, whose constraints are the information structures and incentives of self-interested agents, and whose claims are stated precisely enough to be wrong. Within that frame we developed two propositions. Norm-based governance, in the forms Ostrom documented, is an engineering solution optimal within an operating envelope defined by small-scale information conditions; the historical record shows successful societies replacing each information function that scale destroys with a designed institution that reproduces it. And electoral delegation, operating at scales where those conditions have long since failed, carries the structure of a market with asymmetric information about latent quality — exposed to adverse selection whose intensity should depend, testably, on screening institutions, signaling costs, and the informational environment. We have deliberately claimed less than we suspect. The propositions are stated conditionally; the counterevidence is cited alongside the support; the formal models are posed as problems rather than presented as results. This is not modesty for its own sake. A research program at version 0.2 earns credibility by exposing its load-bearing assumptions, not by decorating them. We conclude with the frame we believe the question deserves: > If governance can be fruitfully modeled as institutional engineering, then existing theories from economics, political science, computer science, control theory, and complex systems may be understood within a common analytical framework. Whether this abstraction yields predictive and practically useful theories remains an empirical and mathematical question — one we intend to pursue, and one on which we invite collaboration and criticism. A companion line of work by the authors develops one concrete design instance — an institutional architecture built around pre-electoral competence certification, audited outcome measurement, and binding performance contracts — as a test article for the framework developed here. We keep it out of this paper deliberately: the framework must stand or fall on its own, and be usable by designers who reach entirely different designs. --- ## References Akerlof, G. A. (1970). The market for "lemons": Quality uncertainty and the market mechanism. *Quarterly Journal of Economics*, 84(3), 488–500. Arrow, K. J. (1951). *Social Choice and Individual Values*. Wiley. Asher, S., Novosad, P., & Rafkin, C. (2024). Intergenerational mobility in India: New measures and estimates across time and social groups. *American Economic Journal: Applied Economics*, 16(2), 66–98. https://doi.org/10.1257/app.20210686 Ashworth, S. (2012). Electoral accountability: Recent theoretical and empirical work. *Annual Review of Political Science*, 15, 183–201. Åström, K. J., & Murray, R. M. (2008). *Feedback Systems: An Introduction for Scientists and Engineers*. Princeton University Press. Besley, T. (2005). Political selection. *Journal of Economic Perspectives*, 19(3), 43–60. Buchanan, J. M., & Tullock, G. (1962). *The Calculus of Consent*. University of Michigan Press. Caselli, F., & Morelli, M. (2004). Bad politicians. *Journal of Public Economics*, 88(3–4), 759–782. Dal Bó, E., Finan, F., Folke, O., Persson, T., & Rickne, J. (2017). Who becomes a politician? *Quarterly Journal of Economics*, 132(4), 1877–1914. Downs, A. (1957). *An Economic Theory of Democracy*. Harper. Dunbar, R. I. M. (1993). Coevolution of neocortical size, group size and language in humans. *Behavioral and Brain Sciences*, 16(4), 681–694. Greif, A. (1993). Contract enforceability and economic institutions in early trade: The Maghribi traders' coalition. *American Economic Review*, 83(3), 525–548. Hardin, G. (1968). The tragedy of the commons. *Science*, 162(3859), 1243–1248. Hayek, F. A. (1945). The use of knowledge in society. *American Economic Review*, 35(4), 519–530. Hurwicz, L. (1973). The design of mechanisms for resource allocation. *American Economic Review*, 63(2), 1–30. Maskin, E. S. (2008). Mechanism design: How to implement social goals. *American Economic Review*, 98(3), 567–576. Milgrom, P. R., North, D. C., & Weingast, B. R. (1990). The role of institutions in the revival of trade: The law merchant, private judges, and the Champagne fairs. *Economics & Politics*, 2(1), 1–23. Myerson, R. B. (2008). Perspectives on mechanism design in economic theory. *American Economic Review*, 98(3), 586–603. Nisan, N., Roughgarden, T., Tardos, É., & Vazirani, V. V. (Eds.). (2007). *Algorithmic Game Theory*. Cambridge University Press. North, D. C. (1990). *Institutions, Institutional Change and Economic Performance*. Cambridge University Press. Olson, M. (1965). *The Logic of Collective Action*. Harvard University Press. Ostrom, E. (1990). *Governing the Commons: The Evolution of Institutions for Collective Action*. Cambridge University Press. Ostrom, E. (2010). Beyond markets and states: Polycentric governance of complex economic systems. *American Economic Review*, 100(3), 641–672. Rawls, J. (1971). *A Theory of Justice*. Harvard University Press. Roth, A. E. (2002). The economist as engineer: Game theory, experimentation, and computation as tools for design economics. *Econometrica*, 70(4), 1341–1378. Roth, A. E., Sönmez, T., & Ünver, M. U. (2004). Kidney exchange. *Quarterly Journal of Economics*, 119(2), 457–488. Scott, J. C. (1998). *Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed*. Yale University Press. Sen, A. (2009). *The Idea of Justice*. Harvard University Press. Spence, M. (1973). Job market signaling. *Quarterly Journal of Economics*, 87(3), 355–374. --- *This is a working paper of Niti Shakha, the research institution of the Neocrates project. Version 0.2 exists to stabilize concepts and invite criticism before any journal submission. Comments, counterexamples, and pointers to literature we have missed are actively solicited.*