PRIORITY FILING - INITIAL DRAFT
A fully revised and formatted version will follow.
Gregory Caldwell Beier
Susarb LLC, Cambridge, Massachusetts
March 19, 2026
From the Prolegomena to the Unification of Economics, Computer Science, Mathematics, and Physics by Gregory Caldwell Beier (forthcoming 2026).
© 2026 Gregory Caldwell Beier. Licensed under CC BY-NC-ND 4.0.
The Disclosure Incentive (Beier, 2026b) is a mechanism. Mechanisms require infrastructure. A proper scoring rule cannot score if no system observes and records the outcomes against which disclosures are evaluated. A track record cannot accumulate if no registry stores and makes accessible the sequence of scored judgments. Disclosures cannot be compared across analysts if no shared ontology defines the constraint domains in which they operate. This paper identifies and analyzes the four components of epistemic infrastructure that must exist before the Disclosure Incentive can function: outcome observation systems, scoring computation platforms, track record registries, and constraint domain ontologies. Each is a public good with network effects - non-rival in use, partially non-excludable, and more valuable to each participant as the number of participants grows. The economics of building this infrastructure are the economics of a coordination problem: the infrastructure is undersupplied by private actors because each component's value depends on the others existing (outcome observation is useless without a scoring platform; a scoring platform is useless without a track record registry; a registry is useless without an ontology that makes disclosures comparable), and the interdependence creates a chicken-and-egg problem that the standard market provision of public goods cannot solve. The paper draws on Frischmann's (2012) framework for infrastructure economics and Ostrom's (1990) institutional design principles to analyze the provision problem and propose a sequencing strategy for building the epistemic substrate on which the Disclosure Incentive depends.
Keywords: epistemic infrastructure, public goods, network effects, outcome observation, scoring platforms, track records, ontology, institutional design, coordination problem
JEL Codes: H41, D85, L86, D02, O33
The companion papers in this series develop the Disclosure Incentive as though the infrastructure it requires already exists. The formal model (Beier, 2026b) assumes that disclosed signals become public upon publication, that proper scoring rules are applied to them, and that track records accumulate. The incentive compatibility proofs (Beier, 2026r) assume that outcomes are observed, that scores are computed, and that principals can access track records when making hiring and delegation decisions. The adversarial analysis (Beier, 2026an) assumes that detection algorithms can be applied to the disclosed data stream.
None of this infrastructure exists at scale. Building it is the practical precondition for everything the companion papers propose, and the economics of building it are distinct from the economics of the mechanism that will operate on it. The present paper is the practical paper in the series - the one that asks what you actually need to build, what it costs, who pays for it, and in what sequence the components must be assembled.
Section 2 identifies the four infrastructure components. Section 3 analyzes their public good properties. Section 4 develops the interdependence problem. Section 5 proposes a sequencing strategy. Section 6 examines funding models. Section 7 states limitations. Section 8 concludes.
Proper scoring rules compare disclosed probabilities to realized outcomes. The scoring is only as reliable as the outcome observation. For financial market predictions (will this company's earnings exceed $X?), outcome observation is largely automated - financial data services report earnings, and the comparison can be computed mechanically. For constraint domain predictions (will this aquifer cross a depletion threshold within seven years?), outcome observation requires physical monitoring infrastructure - sensor networks, satellite imagery, water level measurements, atmospheric monitoring stations - that may or may not exist in the relevant geography.
The quality of the epistemic infrastructure depends first on the quality of outcome observation. An aquifer Scaling assessment scored against unreliable water level data produces unreliable calibration scores, which produce unreliable track records, which undermine the entire incentive structure. The foundation must be solid or nothing built on it is trustworthy.
Outcome observation infrastructure already exists in many constraint domains - weather station networks, the Landsat satellite program, EPA monitoring stations, USGS stream gauge networks - but it was built for scientific and regulatory purposes, not for scoring economic judgments. Adapting existing observation infrastructure to serve the Disclosure Incentive requires standardizing the interface between observation systems and scoring platforms, which is an integration task rather than a construction task in domains where monitoring already operates. In domains where monitoring is sparse (deep aquifer systems in developing countries, ecosystem health in remote regions, infrastructure condition in aging systems), building the observation infrastructure is a prerequisite that may require public investment.
This public-goods funding requirement for the base infrastructure does not contradict the private-incentive transition dynamics developed in the companion paper on adoption thresholds. The information rent on constraint dimensions provides the economic fuel for adoption and participation by asset managers, justifying their internal switching costs. The shared epistemic rails on which they operate - the registries, the scoring platforms, the constraint domain ontologies - are non-excludable natural monopolies that require consortium or public funding to initialize. Private incentive fuels the participation; shared infrastructure must be built first.
A scoring computation platform takes disclosed probability assessments, retrieves the relevant outcomes from observation systems, applies the proper scoring rule, computes the score, and publishes the result. The platform must handle the temporal complexity of the Disclosure Incentive - some outcomes are observed within months, others take years or decades - and must maintain the scoring pipeline across the full range of disclosure cadences and constraint domains.
The computation itself is direct. Applying a Brier score or logarithmic score to a disclosed probability and an observed outcome is a trivial calculation. The engineering challenge is in the pipeline: matching each disclosure to its relevant outcome observation, handling partial outcomes (the aquifer hasn't crossed the threshold yet - is that a non-event to be scored or an incomplete observation to be deferred?), managing the temporal queue of disclosures awaiting scoring, and publishing results in a format that principals and the track record registry can consume.
The platform's design choices have economic consequences. Whether scoring is performed by a single centralized platform (efficient but creates a single point of failure and potential for manipulation) or by multiple competing platforms (redundant but resilient) is an architecture decision with implications for the adversarial robustness that the companion paper (Beier, 2026an) analyzes. Whether the scoring algorithm is open-source (maximizes trust and auditability) or proprietary (enables commercial platform competition) affects adoption incentives and the speed of ecosystem development.
A track record registry stores the cumulative scored history of each discloser - every disclosure, its score, the outcome it was scored against, and the metadata (domain, cadence, confidence level, scoring rule applied) needed to evaluate the record's meaning. The registry must be persistent (records must survive across decades), portable (the record belongs to the agent, not to any platform), and tamper-resistant (historical scores cannot be retroactively altered).
The registry is the infrastructure component most directly analogous to existing financial infrastructure. Credit bureaus maintain cumulative records of borrower behavior that lenders consult when making credit decisions. The track record registry serves the same function for judgment quality - principals consult the registry when making delegation decisions - but with different data (calibration scores rather than payment histories) and different properties (the record is public rather than access-controlled, and the agent owns it rather than the bureau).
Lamport's (1978) work on logical clocks for causal ordering and the subsequent development of cryptographic hash chains (Merkle, 1979) provide the technical foundation for tamper-resistant temporal ordering. A track record entry that is timestamped and hash-linked to its predecessor cannot be retroactively inserted, deleted, or altered without breaking the chain. The infrastructure exists conceptually; building it at the scale required for a global judgment market is an engineering and governance challenge.
Disclosures are comparable only within shared constraint domains. A hydrologist's aquifer assessment and an atmospheric scientist's carbon loading assessment both use the five-field format, but they address different constraints, use different outcome variables, and are scored against different observations. A constraint domain ontology specifies, for each domain, what constitutes a well-formed Scaling assessment, what outcome variables the scoring will use, what observation systems provide the outcome data, and what cadences are appropriate.
The ontology is the most intellectually demanding component to build, because it requires domain expertise (what are the relevant constraint variables in this domain?), information-theoretic judgment (what level of specificity makes disclosures comparable without making the format so rigid that it excludes valid analytical approaches?), and ongoing maintenance (as the constraint environment evolves, the ontology must evolve with it). A poorly designed ontology - one that is too rigid, too vague, or incorrectly specified - degrades the entire system's performance by making disclosures non-comparable or by scoring them against inappropriate outcome variables.
Each component exhibits the defining properties of a public good with network effects.
Outcome observation is non-rival (one user's access to aquifer level data does not reduce another's) and partially non-excludable (data from public monitoring networks is typically available to all). Its value increases with the number of disclosers whose assessments it scores, because a larger scoring base produces more reliable calibration estimates.
Scoring computation is non-rival (computing one score does not consume capacity needed for another, given adequate infrastructure) and can be made non-excludable through open-source platforms. Its value increases with the number of analysts scored and the number of principals who consult the scores.
Track record registries are non-rival (consulting one agent's record does not diminish another's) and most valuable when they cover a large fraction of domain-relevant analysts. A registry that covers only 20% of active analysts in a domain is less useful to principals than one covering 80%, because the principal can evaluate candidates only against the covered population.
Constraint domain ontologies are non-rival (one user's adoption of the ontology does not reduce another's) and exhibit strong network effects - an ontology is useful only to the extent that other participants in the domain use the same one, which makes ontology adoption a coordination game with potential for fragmentation if competing ontologies emerge.
Frischmann (2012) analyzed the economics of infrastructure that shares these properties, arguing that traditional infrastructure (roads, utilities, communication networks) is undersupplied by private markets because of the non-rivalry, non-excludability, and network effects that make private capture of the infrastructure's value difficult. The epistemic infrastructure has the same structure, and the same provision problem applies.
The four components are not independently valuable. Each depends on the others existing for its own value to be realized.
Outcome observation without a scoring platform produces data that no one scores. A scoring platform without a track record registry produces scores that no one accumulates. A registry without an ontology stores records that are not comparable across analysts. An ontology without observation, scoring, and registry infrastructure is a specification that nothing instantiates.
The interdependence creates a coordination problem that compounds the public good provision problem. Not only is each component undersupplied individually (because of its public good properties), but the value of investing in any single component depends on whether the other components will also be built. An investor in outcome observation infrastructure faces uncertainty about whether a scoring platform will emerge to use the data, and the uncertainty depresses the investment's expected return. The circular dependency - each component's value contingent on the others - is a classic chicken-and-egg problem that the standard market provision of individual public goods does not resolve.
Ostrom's (1990) analysis of commons governance is relevant here. Ostrom's central finding is that communities solve collective action problems through institutional designs that sequence investments, establish credible commitments, and create feedback loops that reward early contributors. The epistemic infrastructure requires a similar institutional design: a sequencing strategy that breaks the circular dependency by building components in an order that creates value at each stage rather than requiring all four to exist simultaneously.
The circular dependency can be broken by building on domains where partial infrastructure already exists.
The first stage targets constraint domains where outcome observation is already robust - financial markets (where price and earnings data are abundant), weather-dependent agriculture (where meteorological observation networks are mature), and regulated utilities (where grid performance data is systematically collected). In these domains, the observation infrastructure exists. The investment needed is a scoring platform that connects to existing observation systems, an initial ontology for the domain, and a minimal track record registry.
The second stage adds constraint domains where observation infrastructure exists but is not yet connected to a scoring pipeline - aquifer monitoring networks, air quality stations, satellite-based land use observation. The connection is primarily an integration task: building the data pipelines that link existing observation to the scoring platform developed in stage one.
The third stage addresses constraint domains where observation infrastructure must be built or substantially expanded - deep aquifer systems in regions without monitoring, ecosystem health indicators in data-sparse geographies, infrastructure condition assessment in aging systems. This stage requires the largest investment and the longest timeline, and it is justified by the value that stages one and two have demonstrated.
The sequencing works because each stage creates value that justifies the next. Stage one produces scored track records in domains with existing observation, demonstrating the mechanism's viability and building the track record premium that attracts analysts. Stage two extends the scoring to domains with existing but unconnected observation, expanding the ecosystem. Stage three builds new observation infrastructure in domains where the demonstrated value of stages one and two justifies the investment.
The public good properties of epistemic infrastructure suggest that private provision alone will be insufficient, particularly for stage three. Several funding models are compatible with the infrastructure's economics.
The first is public investment, analogous to the government funding of weather observation networks, geological surveys, and satellite programs. The epistemic infrastructure produces coordination benefits that the individual investor cannot capture, and the case for public funding follows the same logic that justifies public investment in scientific infrastructure.
The second is consortium funding, where the institutional investors who benefit most from the infrastructure - pension funds, sovereign wealth funds, insurance companies whose long-horizon liabilities make constraint information valuable - pool resources to build the platform they collectively need. Frischmann (2012) documents precedents for consortium-funded infrastructure in other domains.
The third is open-source development, where the scoring platform, registry, and ontology are built as open-source projects sustained by the community of users and maintained through the same reputation incentives (Lerner and Tirole, 2002) that the Disclosure Incentive uses for analyst participation. The companion paper on the Disclosure Incentive (Beier, 2026b) develops the structural parallel between open-source software development and the judgment market; the same logic applies to the infrastructure on which the mechanism runs.
The likely outcome is a hybrid: public investment in outcome observation (where governments already operate), consortium investment in the scoring platform and registry (where institutional investors have the strongest interest), and open-source development of the ontology (where domain expertise from multiple communities must converge).
The scoring infrastructure faces an oracle problem that this paper underweights. While the mathematics of proper scoring rules is well understood, the epistemic verification of ground-truth outcomes in complex constraint domains is adversarial, not trivial. When a high-profile analyst's track record capital depends on whether an aquifer crossed a depletion threshold, the outcome data itself becomes contested. Dispute resolution mechanisms, cryptographic verification of observation data, and legal arbitration layers will be required for any operational deployment. The scoring calculation is simple; the institutional machinery that produces the input to that calculation is not.
The paper identifies the infrastructure components and their economics but does not estimate the cost of building them. A cost estimate would require domain-specific engineering analysis (what does it cost to connect the USGS aquifer monitoring network to a scoring platform? what does it cost to build outcome observation for ecosystem health in the Amazon basin?) that is beyond the scope of a framework paper.
The sequencing strategy assumes that partial infrastructure creates demonstrable value that justifies subsequent investment. If the early stages do not produce a convincing track record premium - if principals do not learn to value scored constraint assessments over existing proxies - the investment case for later stages collapses. The sequencing is a bet on the mechanism's viability, and the bet could fail.
The governance of the infrastructure is not addressed. Who controls the ontology? Who arbitrates disputes about outcome observation? Who ensures that the scoring platform is not captured by interests that benefit from specific scoring outcomes? These governance questions are consequential for the infrastructure's trustworthiness and neutrality, and they require institutional design that this paper identifies as necessary but does not provide.
The paper treats the four components as necessary and sufficient. Other infrastructure components - identity verification systems for disclosers, legal frameworks for the safe harbor proposed in the companion paper (Beier, 2026b), educational infrastructure for training analysts in the five-field format - may also be prerequisites that the four-component framework does not capture.
The Disclosure Incentive is a mechanism that runs on infrastructure. The infrastructure does not yet exist at scale, and its construction is a coordination problem with public good properties that private markets alone will not solve. Four components are needed: systems that observe outcomes, platforms that compute scores, registries that store track records, and ontologies that make disclosures comparable. Each is non-rival, network-effect-driven, and valuable only when the others exist alongside it.
The interdependence is the practical barrier to implementation - not the formal model, not the incentive compatibility proofs, not the adversarial defense architecture, but the concrete question of what to build first when everything depends on everything else. The sequencing strategy proposed here breaks the dependency by starting where observation infrastructure already exists, demonstrating value in those domains, and expanding outward as the demonstrated value justifies further investment. The Disclosure Incentive's theoretical architecture is complete in the companion papers. Its practical architecture begins with the epistemic infrastructure this paper describes.
This paper presents results from the Prolegomena to the Unification of Economics, Computer Science, Mathematics, and Physics (Beier, forthcoming 2026). Epistemic infrastructure is the practical foundation on which the Disclosure Incentive (Beier, 2026b) and its formal properties (Beier, 2026r) depend. The author invites engagement from the infrastructure economics, institutional design, and technology policy communities.