From Verifiability to a Productivity Theory of Public Trust Roots
Zhu, Weisha, and Scott Shields September 12,2026
Abstract
Modern societies depend heavily on trust, yet economics usually studies how trust or credit affects transactions, finance, and growth rather than treating the production of trust itself as a productivity problem. In banking, securities markets, and other institutional settings, many key facts cannot be independently verified by outsiders. Internal controls, risk management, audit, regulation, legal liability, capital rules, ratings, and repeated due diligence are therefore used to transform claims into facts that other actors are willing to rely on. These arrangements constitute a system of trust production, and they consume real social resources.
Over the past several decades, computing, databases, the Internet, and automation have sharply reduced the cost of recording, calculating, and transmitting information. Yet the unit cost of financial intermediation has not fallen on a comparable scale. Philippon’s long-run estimates for the United States place the annual unit cost of intermediating one dollar of financial assets at roughly 1.5–2 percent. In this paper, that observation is used as an empirical entry point rather than as an already-solved identification result. We ask whether, in addition to product complexity, risk bearing, entry barriers, zero-sum trading, and rents, verification, audit, compliance, supervision, and professional judgment constitute a distinct class of real social costs that should be separately accounted for.
Verifiability and public trust roots provide a different way to produce trust. They do not make trust free, eliminate future risk, resolve legal judgment, or guarantee that real-world inputs are true. Their promise is narrower: to shift part of the trust that is currently produced through institutional reputation, repeated supervision, and repeated responsibility chains into facts that can be machine-verified, continuously reviewed, replayed, and reused across actors.
The paper therefore treats trust as an independent output and develops a unified accounting framework for trust-production productivity. “Quantification” is used in the economic-accounting sense: specifying inputs, quality-adjusted outputs, reuse, and comparable accounting boundaries. It is not a claim to a natural-science constant, nor does it presuppose that all variables have already been econometrically identified.
The framework uses a counterfactual rule to define the cost of trust production C, a quality constraint Q based on the probability of falsely accepting a key fact, an effective-output measure S_eff that counts only trust services that support real economic, legal, or responsibility-bearing action, and a reuse rate R that measures how often an already verified fact is actually relied upon by independent relationships. It then adds the shared and usage costs of public trust roots. Public data from Philippon, the U.S. Bureau of Labor Statistics, the Conference of State Bank Supervisors, the PCAOB, and corporate filings are used to map the framework into observable quantities. USDC is used to construct a first partially identified industry account, not to claim that a complete industry productivity estimate has already been obtained.
Keywords: trust-production productivity; fact-based trust; verifiability; trust reuse; public trust root; USDC
1. A Long-Neglected Question
What are supervision, audit, regulation, and legal responsibility in finance jointly producing? One possible answer is: trust.
In this paper, “producing trust” means transforming a claim or state that outsiders cannot directly confirm into a sufficiently credible fact on which they are willing to act. That transformation consumes resources. Once this is recognized, productivity becomes a meaningful question, analogous in structure to labor productivity.
What does “trust” mean here?
The Chinese term xinyong (信用) spans credit, trust, and credibility. The object of this paper is narrower than any of those terms in ordinary language. It is a key fact or state that has reached a level of credibility at which an external actor is willing to take an action with economic, legal, or responsibility consequences. Broader forms of trust based on emotion, habit, identity, authority, reputation, or social ties lie outside the core accounting object unless they are attached to verifiable factual states.
The paper therefore uses the term fact-based trust: an actionable state of confidence created when key facts have passed a specified process of verification. It is not a guarantee of future outcomes. It is a reduction in uncertainty about facts that exist now or have already occurred. Four activities must be separated conceptually: information processing, which records, computes, transmits, and stores facts; trust production, which confirms key facts and makes them independently reviewable; risk bearing and risk pricing, which address future uncertainty; and institutional judgment, which interprets rules, allocates responsibility, protects users, and handles exceptions.
In practice these activities overlap. For accounting purposes, however, only the real resources that would fall if a key fact became directly, continuously, and independently verifiable should be counted as trust-production costs.
Consider a bank that reports assets of RMB 10 billion. The number itself is initially only a claim. To make it actionable, the bank keeps records and controls, management accepts responsibility, internal audit tests the process, external auditors obtain evidence, regulators receive reports and supervise, legal rules attach liability to false statements, and counterparties or investors may perform further due diligence. Trust is therefore not costless. It is a social product that requires real resources.
The idea that institutions can “produce trust” is not itself new. Neu (1991), for example, used the phrase institutional production of trust to describe how securities regulation, the accounting profession, and market rules jointly create the trust needed for new equity issues. The question advanced here is different: can the production of fact-based trust be given an accounting boundary, a quality standard, an output measure, a reuse measure, and a productivity measure?
Like GDP or labor productivity, trust-production productivity should not be expected to yield a unique and permanently exact constant. Economic statistics depend on classification, accounting conventions, quality adjustment, valuation, and evolving measurement methods. Their usefulness comes from transparent definitions, stable accounting rules, reproducibility, and consistent comparison. The same methodological standard should apply here.
The paper’s academic position is therefore at a different level from theories that explain particular mechanisms of transactions, contracts, monitoring, verification, institutional trust, or consensus security. Those literatures explain components of the process. This paper defines the higher-level accounting object served by those mechanisms: how fact-based trust reaches a required quality, becomes actionable to outsiders, and can be measured in terms of cost, effective output, and reuse.
| Literature / theory | Primary object | Relation to this paper |
| Transaction-cost and institutional theory | How transactions are organized; how institutions reduce uncertainty and transaction costs. | Explains the institutional environment in which trust is formed, but does not treat fact-based trust itself as a productivity-accounting output. |
| Agency, verification, and monitoring theory | Costs and incentives of monitoring, guarantees, costly state verification, and delegated monitoring. | Supplies important local components of C, but the unit of analysis is usually a particular agency problem or financial contract. |
| Institutional and social trust research | How institutions generate trust and how trust affects exchange and aggregate productivity. | Shows that trust can be institutionally produced and economically valuable; this paper asks how the production of fact-based trust itself can be accounted for. |
| Blockchain verification and shared security | Verification cost, networking cost, shared security, and settlement infrastructure. | Explains why particular technologies may reduce local costs; this paper places those costs inside one account with Q, S_eff, R, and root-level costs. |
| Economics of Nakamoto trust | Security budgets, attack incentives, and scale constraints of permissionless consensus. | Provides a key constraint on public-root security budgets and scale, but does not measure cross-industry effective trust services. |
| This paper | Production, quality, cost, reuse, and public roots of fact-based trust. | Builds industry-level, publicly calibratable, third-party-reproducible productivity accounts using C, Q, S_eff, R, and public-root costs. |
2. Philippon’s Observation: Why Did the Information Revolution Not Produce a Comparable Fall in Financial Intermediation Costs?
Philippon’s long-run evidence offers a useful empirical entry point. The U.S. financial sector changed greatly in size and composition, but after adjustment the unit cost of financial intermediation remained around 1.5–2 percent per dollar of intermediated assets per year. This is not 1.5–2 percent of GDP; it is roughly 1.5–2 cents of annual intermediation cost for each dollar of intermediated assets.
That observation is striking because the cost of recording, copying, calculating, and transmitting information fell by orders of magnitude over the same broad period. Philippon himself discusses multiple explanations, including changing products and clients, zero-sum trading, entry barriers, regulation, and economic rents. The unit-cost series therefore cannot simply be relabeled “trust cost.”
A useful conceptual decomposition is to separate necessary financial functions, pure information processing, trust production, and rents or other structural costs. Philippon measures the total. The present paper asks whether verification, supervision, compliance, audit, internal control, and professional judgment form a distinct component within that total. The task here is to define the component and its accounting boundary. Estimating its eventual share in the Philippon residual is a later empirical exercise.
This leads to a testable conjecture: information-processing productivity improved dramatically, while the productivity of producing trustworthy facts may not have improved on the same scale. Machines have become extraordinarily good at calculation. The expensive question may no longer be whether 2 + 2 = 4, but where the first “2” came from. Do the assets exist? Was the transaction authorized? Are liabilities omitted? Is collateral reused? Was a model manipulated? Did the regulator receive complete data?
Initial evidence: one side falls, the other remains resource-intensive
Public data already show two opposite directions, even though they do not yet identify the Philippon residual. First, routine information-processing work has been increasingly automated. Historical BLS material reported that clerical and administrative support occupations accounted for about 68.5 percent of commercial-banking employment in 1987. In the broader 2023 category of credit intermediation and related activities, office and administrative support occupations accounted for about 39.8 percent. Because the industry and occupational classifications differ, these figures are used only as directional background, not as a same-definition time series.
Second, compliance, audit, and professional judgment continue to consume substantial resources. CSBS research pairing 2015–2024 survey data with Call Report information found that the smallest community banks attributed roughly 11–15.5 percent of personnel expense to regulatory compliance. In the 2025 survey, respondent banks on average attributed 13 percent of personnel expenses, 36 percent of accounting and audit expenses, and 29 percent of consulting and advisory expenses to compliance; personnel costs represented about 58 percent of total compliance spending. The denominators differ, so the 13 percent and 58 percent figures are not inconsistent.
PCAOB data point in the same general direction: registered accounting-firm staffing increased over the past decade, and staffing normalized by issuer revenue also rose. The PCAOB cautions that these personnel may perform work beyond issuer audits, so the series should be treated as a directional proxy for audit-resource intensity rather than a direct measure of trust-production cost.
The strongest conclusion available at this stage is therefore modest: routine information processing has a clear long-run automation trend, while verification, control, and assurance continue to absorb significant resources. How much of those resources meet the counterfactual cost boundary defined below, and how much of Philippon’s cost stickiness they explain, remain open empirical questions.
3. The First Accounting Boundary: What Counts as a Cost of Trust Production?
Without a cost boundary, trust-production productivity cannot be computed. The simplest test is counterfactual: if a key fact were already directly, continuously, and independently verifiable by outsiders at low cost, how much of the expenditure would remain? If none of it would fall, the activity is usually outside C. If part would fall, only that part belongs in C. If the activity would largely disappear, it is strongly related to trust production.
First: the product boundary
Banks, regulators, Bitcoin, and Ethereum do not produce identical products. Banks allocate funds, transform maturity, provide payments, and bear risk. Regulators protect consumers, set rules, manage crises, preserve financial stability, and enforce law. Bitcoin does none of these things. A valid productivity comparison must therefore isolate a common product: making a key fact or state sufficiently credible that outsiders can act on it. That common product is fact-based trust.
Second: the cost boundary
Not every bank or regulatory cost is a trust-production cost. C includes real resources used because key facts cannot otherwise be socially accepted at the required quality. Depending on the setting, this may include parts of internal control, risk management, internal and external audit, regulatory inspection, compliance verification, legal attestation, factual investigation, and repeated due diligence. Ordinary customer service, funding cost, and marketing are not included merely because they occur inside a bank.
Third: social cost is not the same as monetary payment
A billion-dollar fine is not automatically a billion dollars of social resource loss. The relevant real costs may include investigation, litigation, enforcement, remediation, resource misallocation, and actual economic losses. Likewise, a blockchain miner’s dollar-denominated block reward is not automatically equal to the social cost of the network. Energy, hardware, capital, depreciation, networking, and labor are the relevant real-resource inputs.
Fourth: quality must be held constant
A system that spends 1 unit to reach 70 percent reliability cannot be judged more productive than a system that spends 10 units to reach 99.999 percent reliability merely because it is cheaper. Productivity comparisons require a minimum quality threshold.
Q_i(a) ≥ Q₀,i ⇔ p_e,i(a) ≤ ε_i
Here i denotes the industry or trust product and a the production architecture. Q_i(a) is the achieved trust quality, Q₀,i the minimum required quality, p_e,i(a) the probability that a key fact is falsely accepted as true, and ε_i the maximum tolerated false-acceptance probability. Writing Q = 1 − p_e is an idealized representation; it does not imply that every industry can already estimate a single precise probability.
Quality measures need not be identical across industries, but comparisons must be framed around the same underlying question: under the relevant rules and time horizon, how likely is it that a key fact is wrongly accepted? For public blockchains, long-run absence of observed reversals after deep confirmation is evidence of stability, but it does not imply a theoretical error probability of zero. The relevant comparison is always between architectures serving the same trust product at a comparable quality threshold.
4. Three Boundaries of Non-Substitutability: What Verification Cannot Solve
First, a public trust root cannot make an off-chain input true. Bitcoin can prove that a commitment existed by a certain point in history. Ethereum can prove that an on-chain state transition followed protocol rules. Neither can by itself prove that gold is physically in a vault, goods were delivered, a person is who they claim to be, or a sensor was not manipulated. Industry-specific mechanisms must establish the truth of real-world inputs before or while they enter the verification structure.
Second, verification of the past does not eliminate future risk. Ten years without default do not prove that an institution will never default. A fully reserved stablecoin today may not remain fully reserved tomorrow. Verifiability reduces uncertainty about facts; it does not remove uncertainty about the future.
Third, verification does not eliminate law or human judgment. Protocol upgrades, crises, forks, quantum threats, social conflict, proportionality, exceptions, and appeals can require institutional or human decisions. Verification can reduce disputes about facts without eliminating normative judgment. Final responsibility remains human.
5. Trust-Production Productivity
If fact-based trust is a product, productivity can be defined. The central expression is an accounting identity analogous to “qualified output divided by input,” not an already-estimated behavioral production function.
P_i(a) = S_eff,i(a) / C_i(a) subject to: p_e,i(a) ≤ ε_i (1)
P_i(a) is the productivity of architecture a for product i; C_i(a) is its real social cost; S_eff,i(a) is the effective trust service produced at or above the quality threshold. The ratio is comparable only for the same product and a common or equivalent quality requirement.
S_eff excludes technical activity that does not produce real external reliance: raw API calls, repeated anchoring, spam transactions, circular self-use, or nominal activity without economic, legal, or responsibility consequences. The plain-language interpretation is: at a specified quality, how many effective trust relationships can one unit of real social resource support?
Quality may be represented by p_e in theory, but empirical work may have to report a vector of observable proxies Z_Q instead: time coverage, process coverage, sampling share, material misstatement, false acceptance or omission, state reversal, or inability to replay a process. Heterogeneous proxies should not be averaged into a cosmetically precise single Q.
The units of S_eff may be industry-specific. A bank may measure relationships supported by trustworthy asset and liability states; a securities market may use assets or transactions supported by trustworthy disclosure and custody states; a stablecoin may use circulating value backed by reliable reserve states; an AI system may use tasks whose authorization, execution, and responsibility chain have been verified. The theoretical variables are common; the measurement units need not be.
6. Trust Reuse: An Underestimated Variable
Productivity depends not only on how much it costs to produce a trustworthy fact once, but also on how many independent relationships can rely on the result. If a verified fact is used by only one actor, it generates one trust service. If 100 independent economic relationships can rely on the same verified result, the same verification supports more effective trust output.
R_i = U_dep,i / D_ver,i (2)
U_dep is the deduplicated number of effective dependencies on verified facts or proofs: distinct legal entities, contracts, or responsibility relationships that actually act on the proof. D_ver is the number of independent facts or proofs generated in the same period that meet Q₀. R = 1 means that each qualified proof supports one effective use on average; R > 1 indicates factual trust reuse.
A verifiable counting rule for R requires an auditable dependence record. At minimum, each record should identify proof_id, relying_entity_id, relation_id, and validity_window. Repeated queries, page views, API calls, on-chain reads, or internal forwarding by the same legal entity during the same validity period count once. A single proof may support multiple distinct contracts or responsibility relations, but the relation must remain traceable for deduplication and review.
Where the economic importance of relationships differs substantially, a value-weighted reuse statistic R^V may be reported separately. Count-based R answers how many independent relationships rely on an average qualified proof. Value-weighted R^V answers how much economic value is supported. The two have different units and should not be collapsed into one number.
Traditional systems already contain reuse. Delegated monitoring is itself a form of reuse, and large audit firms, regulators, and rating agencies reuse professional capability. The structural difference is that traditional systems mainly reuse institutional capability rather than already-verified facts. Each institution still reacquires data, rechecks facts, forms its own judgment, issues its own report, and assumes responsibility. That is repeated trust production.
7. Reusing Data Is Not the Same as Reusing Trust: From Once-Only to Verifiable Reuse
Digital government has already attacked the problem of repeated data submission. The EU Once-Only Principle seeks to ensure that citizens and firms submit the same information only once, after which authorized agencies can lawfully reuse it. Early European Commission estimates suggested that EU-wide implementation could produce net annual savings of up to roughly EUR 5 billion.
But data reuse is not yet trust reuse. In a conventional digital system, data are collected once and reused by authorized institutions. In a verifiable system, a fact is converted into an independently checkable proof, and multiple actors can verify or rely on that proof subject to access rules. The shift is from repeated submission to single collection, and then from repeated institutional checking to one qualified proof with multiple independent reviews.
8. Public Trust Roots: Extending Trust Production Beyond the Institution
Traditional systems can achieve scale economies inside institutions. A public trust root creates a different possibility: cross-institution, cross-project, and potentially cross-industry reuse of a common verification infrastructure.
“Public” here means openly verifiable and shareable across actors. It does not imply that a public trust root is a pure public good in the strict economic sense. Blockspace, data availability, verification capacity, privacy, access, governance, and security budgets can all create congestion, exclusion, or marginal costs. Public trust roots are better treated as shareable verification infrastructures whose scale properties must be determined by the relationship between effective trust services and total social cost.
Historical proofs may be close to non-rival in review, while capacity and data availability can be congestible, access can be rationed by fees or rules, and security and governance behave more like shared infrastructure. Samuelson, Buchanan, and Ostrom therefore provide useful reference points for decomposing cost layers, but no single textbook public-goods category should be imposed on the entire system.
At the most basic level, the unit cost of the public root is:
C_unit(U_root) = C_root(U_root) / U_root (3)
U_root is the effective trust-service volume actually supported by the public root, and C_root(U_root) is the total real social cost of maintaining those services. Equation (3) is an accounting ratio, not the full cost function. If U_root grows faster than total root cost while quality is maintained, unit cost falls. If total cost rises at the same or a faster rate, scale economy does not arise automatically.
Four conditions are therefore central: quality must not fall as use expands; security and governance costs may rise but should grow more slowly than effective service volume; the marginal costs of data, proof generation, integration, privacy, and access must remain manageable; and the migration, governance, and systemic-failure externalities created by a shared root must remain below the repeated trust-production costs it avoids.
A necessary counterargument: does a public blockchain really have scale economies?
Budish (2025) reaches a result that points in the opposite direction for pure Nakamoto trust: permissionless consensus must continually pay a security budget, and the required budget may rise with attackable value. The result cannot be ignored. The difference lies in the object being measured. Budish asks what security expenditure is needed for permissionless consensus to protect economic value. This paper asks how the comprehensive social cost per effective trust service changes when a public root is repeatedly reused by multiple applications and institutions. If root-level security budgets rise in proportion to or faster than the effective value of trust services, the scale economy posited here weakens or disappears.
9. Ethereum Layer 2 as a Real-World Experiment in Trust Reuse
Ethereum Layer 2 systems illustrate shared infrastructure. If every application built an independent blockchain, each would need its own validator set, security incentives, governance, nodes, historical records, and economic security. Ethereum’s own institutional materials describe a “no bootstrapping” benefit: an L2 can inherit the existing security and decentralization infrastructure rather than building a new validator set from scratch.
The L2 still pays for execution, ordering, proofs, data, development, integration, and use of Ethereum. Public trust infrastructure does not eliminate cost. It changes which layers must be repeatedly rebuilt. A second scale effect comes from aggregation: many lower-level transactions or facts can be compressed into a proof or data commitment submitted to L1. The result is a horizontal scale economy across independent systems and a vertical scale economy through aggregation of lower-level facts.
10. Bitcoin and Ethereum as Two Public Verification Modes
Bitcoin and Ethereum can both serve as public trust roots, but their measurement objects need not be identical. Bitcoin’s distinctive strength is durable public historical ordering: whether a commitment existed, when a record appeared, and whether history has been rewritten. This makes it useful as a broad historical trust root for finance, AI responsibility chains, enterprise records, public documents, identity and authorization proofs, supply chains, voting records, and other historical commitments.
Ethereum adds programmable state, contract execution, asset-state transitions, and final settlement. It is therefore particularly suited to financial trust and rule-execution functions. Its long-run value should not be reduced to a single transaction’s gas fee; the more relevant question is how much effective economic activity a public financial trust infrastructure can support per unit of social cost.
The deeper point is not that future systems must depend on these two specific networks. They demonstrate two independently verifiable modes: public historical-fact verification and public rule/state-transition verification. If Bitcoin is used to support more complex programmable or off-chain business logic, a separate verification layer must perform those industry-specific checks and then anchor the key proof into the public root.
11. Public Roots Can Be Shared; Industry Verification Must Remain Industry-Specific
Trust-production productivity must initially be measured by industry because the critical facts differ. Banks must verify identity, authorization, transactions, assets, liabilities, reserves, and accounting. Supply chains must verify origin, ownership, transport, inspection, and delivery. AI systems must verify who authorized a task, what inputs were used, which model and version executed it, what output was produced, whether final human approval was required, and who bears responsibility.
A plausible architecture therefore has three layers. The industry-fact layer specifies which facts matter. The industry-verification layer defines how those facts become evidence, how they are checked, replayed, and handled when exceptions occur. The public-trust-root layer anchors the proofs, states, or hashes that most need public finality. Business can remain efficiently off-chain while critical facts become independently reviewable and selectively anchored to the public root. Centralization itself is not the core problem; unverifiable centralization is.
12. From Trust Roots to Trust Transmission
A public trust root cannot independently produce all real-world trust. It does not automatically know a bank’s assets, a company’s inventory, an AI authorization, a logistics delivery, or a legal responsibility. Industry-specific structures must first create verifiable facts, after which key proofs can be transmitted to a public root.
The full chain is: real-world fact → industry verification → verifiable evidence → trust transmission → public trust root → multi-party review and reuse. This avoids two extremes: leaving everything inside unverifiable centralized black boxes, or placing every business process, datum, and computation on a public chain. A high-productivity architecture is more likely to combine specialized business processing, industry-specific verification, and public final trust infrastructure.
13. Why a Public Trust Root May Produce Large Savings
Traditional trust production is often institution-specific: banks, securities firms, platforms, and future AI-agent networks each build their own controls, verification, audit, authorization, logging, and responsibility structures. They can reuse professional capability, but they have limited ability to reuse another organization’s already-completed verification of facts.
The economic role of a public trust root is to move the most general layer of final verification infrastructure from “rebuild separately” to “maintain jointly and call when needed.” Industry systems still retain their own factual verification and responsibility structures, but they need not duplicate the same level of public historical and consensus infrastructure.
Potential savings therefore come from three sources: reducing duplicated public verification infrastructure, increasing the reuse of already-produced factual trust, and aggregating lower-level facts or proofs so that root-level cost is spread across more economic activity. The savings are conditional. If the new architecture adds more integration, proof, data, access, and governance cost than the repeated trust production it replaces, the shared infrastructure can itself become an expensive new intermediary.
14. Why AI Turns Trust Production into an Infrastructure Problem
AI raises cognitive productivity: it can read, compare, reason, recommend, and increasingly execute. But AI does not inherently know whether its inputs are true. An agent can optimize a portfolio or choose a loan only if it can rely on the truth of account balances, reserves, liabilities, authorization, and execution history.
Recent technical and industry developments point in this direction. Mastercard’s 2026 Verifiable Intent work framed authorization in agentic commerce as something that must become provable rather than merely assumed. MITRE’s 2026 continuous remote attestation framework similarly argued that being trusted at startup does not guarantee continued trustworthiness during operation. Work on proof of execution treats AI authorization, execution paths, history integrity, and replayability as jointly verifiable objects.
The underlying principle is simple: key facts should be verifiable, key processes replayable, and key judgments accompanied by reasons that can be reviewed. As machines become more autonomous, independently verifiable architectures become more important because machines cannot become ultimate trust anchors merely by possessing a “personality” or reputation. A plausible pattern is machine execution → machine verification → replayable process → public trust root → final human responsibility.
15. From Transmitting Data to Transmitting Proof
Traditional trust production often transfers underlying data to a trusted institution, which inspects the data, forms a conclusion, and assumes responsibility for the opinion. A verifiable structure attempts to move part of that process toward data or state → verifiable proof → independent multi-party review → fact-based trust. The raw data do not disappear; what changes is how outsiders obtain confidence in them.
Catalini and Gans emphasize verification and networking costs. This paper adds another distinction: reducing the cost of a single verification does not by itself change the production structure of trust. Repeated trust production falls only when qualified proofs can actually be reused by multiple independent relationships.
16. How This Framework Differs from “Trustware”
Consensys has used “Trustware” and the “industrialization of trust” to describe public blockchains as digital trust infrastructure. The framework here is narrower and more accounting-oriented. It uses a counterfactual rule to identify which real resources are devoted to transforming key facts into actionable trustworthy states, and it measures cost C, effective trust services S_eff, and reuse R under a common quality threshold.
Calling every legal, insurance, audit, regulatory, and institutional expenditure a “trust cost” would invite double counting and over-attribution. The accounting boundary here includes only the resources that would decline if the key fact became directly, continuously, and independently verifiable. Trust-production productivity is therefore meant to be an empirically contestable accounting concept, not a promotional measure.
17. Minimal Accounting Framework
Units and observable proxies
In applied work, C is typically expressed as the monetary value of real resources consumed over a period. S_eff may be measured as qualified effective relationships per period, or as an economic value scale adjusted under a consistent method. P therefore measures how much effective trust service is supported by each unit of real social cost. Different industries should not directly add heterogeneous S_eff units unless a common value weighting and deduplication method has been established.
| Symbol | Theoretical meaning | Measurement / industry-specific interpretation |
| P_i | Industry trust-production productivity | Effective trust services supported per unit of real social cost at a given quality threshold. Units depend on the industry definition of S_eff. |
| C_i | Total social cost of trust production | Real-resource value per period. Include only the portion that would fall under the counterfactual of direct verifiability. |
| S_eff | Effective trust services | Must correspond to actual economic, legal, or responsibility-bearing reliance by outsiders; exclude spam, repeated anchors, and purely technical calls. |
| p_e / ε | False-acceptance probability / tolerated maximum | Estimate p_e where possible. Otherwise report a proxy vector Z_Q, clearly separating process and outcome proxies; do not average heterogeneous proxies into a single Q. |
| R | Trust reuse rate | How many independent effective relationships actually rely on the same verified fact or proof, after deduplication. |
| P_root | Public trust-root productivity | Effective trust services supported by the root per unit of total real social root cost at the required quality. |
| U_root | Effective trust services of the public root | Must not be equated with raw on-chain transaction count; use real economic reliance, protected value, or a standardized service measure. |
| C_root | Total social cost of the public trust root | All real resources required to support U_root. Expanded in Equation (4). |
| Cost decomposition | Components of C_root | Shareable infrastructure, security, governance, data/proof/access/integration, privacy, and expected failure externalities; do not equate transaction fees or miner rewards with social cost. |
The framework uses two core productivity relationships. For a specific industry or trust product:
P_i(a) = S_eff,i(a) / C_i(a) subject to: p_e,i(a) ≤ ε_i (1)
For a public trust root:
P_root(U_root) = U_root / C_root(U_root) subject to: p_e,root ≤ ε_root
Three quantities must remain distinct. S_eff is the effective output of an industry or trust product. U_dep is the deduplicated dependency count used to compute R. U_root is the effective trust-service volume actually supported by the public root. They belong to different accounts and cannot be substituted for or directly added to one another.
18. USDC: A Partially Identified Account of Two Trust-Production Architectures
USDC reserve trust: a partially identified digital-finance account
USDC provides a relatively clear trust product for illustration. Its on-chain issuance, transfer, and burning states can be directly reviewed on public networks, but the remaining off-chain question still requires external trust: does Circle continuously hold eligible dollar reserve assets sufficient to cover circulating USDC? The product boundary in this section is therefore “eligible reserve assets remain at least as large as circulating USDC,” not Circle’s overall business operations.
The traditional trust-production resources relevant to that product include reserve and liability data acquisition and reconciliation, internal controls, external audit or assurance, regulatory reporting and inspection, repeated counterparty or platform due diligence, and other verification resources caused by opacity. Public filings reveal only part of these costs, so the account remains partially identified.
The accounting period is 2025. Circle reported average USDC circulation of USD 64.870 billion and year-end circulation of USD 75.266 billion. Circle also reported weekly disclosure of reserve holdings and monthly third-party assurance. Annual fee ratios below therefore use average circulation as the denominator to preserve period matching.
Circle’s 2026 proxy statement reported 2025 Deloitte Audit Fees of USD 5,874,885 and Audit-Related Fees of USD 1,565,583, for a combined USD 7,440,468. Relative to average circulation, Audit-Related Fees equal about 0.241 basis points, while Audit plus Audit-Related Fees equal about 1.147 basis points.
Neither ratio is the cost of producing USDC reserve trust. Audit Fees include annual financial-statement audit, quarterly review, statutory subsidiary audits, securities offerings, and other regulatory work. Audit-Related Fees include service-organization control reports, attestation services, and subsidiary regulatory filings. The USD 1.566 million amount is therefore only a broad basket that contains reserve assurance; the actual reserve-assurance cost is a subset that public disclosures do not separate. The broader USD 7.440 million amount reflects a wider external-professional-services envelope and cannot be divided by USDC circulation and relabeled C_old for the reserve product.
| Accounting item | 2025 publicly observable value | Measurement status / boundary |
| Trust product | Eligible reserve assets ≥ circulating USDC | Boundary fixed; excludes payment, marketing, liquidity, and other issuer functions. |
| S_eff scale proxy | Average circulation: USD 64.870bn; year-end: USD 75.266bn | Annual fee ratios use average circulation. This is a value-scale proxy, not transaction count. |
| Q / Z_Q quality proxies | Reserve-holdings disclosure weekly; third-party assurance monthly | Process/frequency proxies, not p_e. They cannot be collapsed into a single Q; a direct false-acceptance estimate is still unavailable. |
| Audit-Related Fees | USD 1,565,583; about 0.241 bp of average circulation | Includes SOC reports, attestation services, and subsidiary regulatory filings; only a broad basket containing reserve assurance, not complete C_old. |
| Audit + Audit-Related | USD 7,440,468; about 1.147 bp of average circulation | A broader professional-services envelope containing many costs unrelated to this trust product; not C_old. |
A Tether-style reserve attestation provides a useful boundary contrast: proving that reserves are sufficient is a narrower trust product than establishing the full external credibility of the issuer as a company. The two should not be assigned the same cost denominator.
A verifiable architecture should therefore not be compared as a “reserve snapshot” versus an audit report. The relevant object is the full factual chain from minting to redemption: user funds enter → a bank or regulated intermediary receives the funds → eligible reserve assets are acquired → stablecoin liabilities are issued → reserve assets and liabilities remain continuously matched → redemption occurs → assets are released or sold → tokens are burned → the process remains replayable.
Off-chain banks and custodians would still handle real funds and real assets. An industry verification layer would convert deposits, asset purchases, issuance, reserve-liability matching, redemption, and burning into reviewable evidence, while key proofs or hashes could be anchored to a public trust root. Regulators would receive the data permitted by law and be able to replay the relevant process.
The structural comparison with periodic sampling is conditional but meaningful. Holding source truth and verification scope constant, continuous, comprehensive, replayable verification has greater time coverage, process coverage, and reviewability than periodic sampling, and can expose anomalies earlier. This does not mean off-chain inputs become true merely because they are verified by a machine. If source data are false, an automated system may verify false data faster. Source truth must still be established by banks, custodians, industry verification mechanisms, and legal responsibility.
The first purpose of the USDC example is therefore not to claim that a verifiable architecture is already cheaper. It is to place two methods of producing the same trust product into one account. The traditional architecture relies more heavily on periodic evidence collection, assurance, and institutional responsibility. A verifiable architecture embeds key facts into continuous verification and replay. The traditional side is only partially observed, and the new side still lacks complete implementation-cost data. Cost superiority remains an empirical question; differences in coverage and verification speed can be evaluated earlier under explicit conditions.
From partial identification to interval estimates: what data are still missing?
| Side | Major cost items still to identify | Purpose |
| Traditional trust production | Reserve-specific assurance fee; internal-control and reconciliation labor directly tied to reserve facts; regulatory inspection resources; repeated due diligence by exchanges, custodians, and institutions | Decompose observable professional fees into a reproducible interval for C_old. |
| Verifiable trust production | Bank/custodian data interfaces; source-fact validation; verification-layer operation; proof generation; allocation of public-root cost; exception review; privacy, governance, and necessary compliance responsibility | Build complete C_new from real implementation records rather than substituting a single on-chain fee for total cost. |
| Common quality condition | Time and process coverage; sampling versus full population; false acceptance or omission; anomaly-detection speed; replayability | Compare architectures only for the same trust product and similar Q threshold. |
Only when these data become available should the USDC account be advanced from partial identification to interval estimates for C_old and C_new. The framework does not assume ex ante that one architecture must be cheaper; it first requires a reproducible comparison for the same product, quality condition, and time period.
19. From Industry Accounts to a Social Trust-Production Account
With sufficient data, one could construct separate trust-productivity accounts for banking, securities, insurance fact verification, stablecoins, and AI. Only after those accounts become stable and reproducible should a broader social account be attempted.
At the social level, the question would be how many real resources a society spends each year to confirm facts, verify authorization, prove assets and liabilities, supervise behavior, and create publicly actionable trusted states—and how much effective economic activity those trust services support.
Unified variables do not imply unified units. USDC circulation, bank assets, and AI tasks cannot simply be added. Nor does Q₀ need to be converted into one cross-industry number at the current stage. Q₀ first serves as a within-product quality constraint. A social account would later require a common valuation scale, deduplication, quality adjustment, and cross-industry weighting.
This is analogous to the historical development of national accounting and productivity statistics. Measurement does not become meaningless because accounting conventions, quality adjustments, and estimation errors are debated. But the more aggregate the measure, the less plausible it is to expect natural-science precision. A social trust-production account should therefore remain a long-run research agenda; in the near term, the practical objective is a family of reproducible industry satellite accounts.
20. Failure Conditions, Governance, and Political Economy
Trust-production productivity must include more than normal-period average cost. When a public root becomes common infrastructure, governance, incentives, and failure externalities enter the total social cost. The predicted productivity advantage can shrink or disappear under at least seven conditions.
First, real-world inputs or oracles are captured. A public root can make evidence hard to rewrite after entry; it cannot turn a false input into a true fact. Second, proofs are not truly reusable because institutions still require the same raw data, formats, and responsibility process. Third, governance or verification power becomes excessively concentrated, reintroducing single-point trust and rents. Fourth, security budgets rise as fast as or faster than the value or responsibility protected. Fifth, capacity, data availability, privacy, proof generation, and integration costs become too large. Sixth, technical finality does not connect to legal finality, cross-border responsibility, exception handling, and appeal. Seventh, a highly reused public root creates new systemic concentration risk.
These are not external caveats. They are conditions under which the theory can be empirically rejected. For the same trust product and a comparable quality threshold, if a verifiable architecture fails to increase effective output, reuse, or verification coverage, or if its total social cost is not lower than the repeated trust production it replaces, the claimed productivity advantage does not hold.
The total cost of a public trust root can be decomposed as:
C_root(U_root) = F₀ + B(U_root) + G(U_root) + C_use(U_root) + E(U_root) (4)
F₀ is shareable base infrastructure; B(U_root) is the security budget as scale and protected value change; G(U_root) is governance and upgrade cost; C_use(U_root) covers data, proofs, access, and integration; E(U_root) is the expected externality of concentrated failure and related risks. Equation (4) expands the total cost in Equation (3). Scale economy exists only if total cost grows more slowly than effective trust services over the relevant range while the quality constraint remains satisfied.
21. What the Theory Is Trying to Explain
The Industrial Revolution reduced the cost of material production. The information revolution reduced the cost of recording, calculation, and communication. AI is reducing the cost of cognition. Yet another large social cost has remained largely hidden inside institutions: the cost of making facts worth relying on.
Historically, societies have produced trust through personality, institutions, professional qualifications, regulation, law, audit, and responsibility. These sources will not disappear. Verifiability adds another source: verification of a verifiable fact can itself produce trust.
When key facts continuously generate machine-verifiable evidence, when a qualified proof can be reused by multiple independent economic relationships, and when the security and governance cost of a public root grows more slowly than the effective trust services it supports, trust production acquires a productivity structure that differs from repeated institutional checking.
The economic importance of Bitcoin and Ethereum in this framework is not a token-price narrative. They show that high-quality public trust can be infrastructural: a durable public verification structure can be referenced by other systems. Whether its trust-production productivity is in fact higher than traditional alternatives depends on effective relationships supported, quality achieved, real resources spent on the root and its use, and whether reuse savings exceed governance and systemic-failure externalities.
Conclusion: The Economic Expression of Verifiability
The verifiability framework began with three simple principles: key facts should be verifiable; key processes should be replayable; key judgments should state reasons that can be reviewed. Following these principles through transparent banking, off-chain business with public anchoring, public trust roots, and AI responsibility chains leads to a more basic question: why does society need verification? Because when facts cannot be directly accepted, trust must be produced.
Trust production consumes real resources. Once that is recognized, one can ask about its cost, quality, reuse, and productivity. Public trust roots then change the question from “who should we believe?” toward “which facts can be independently verified, replayed, and reused, and at what social cost?”
Traditional society often produces the same trust repeatedly. Verifiable architectures attempt to convert part of that repeated production into independently checkable and reusable evidence. The research question is not whether this transformation always succeeds, but under what industry conditions, quality thresholds, cost structures, and governance arrangements it increases productivity.
This does not mean that machines replace trust, that Bitcoin or Ethereum solve all social problems, or that law becomes unnecessary. Future risk remains. Real-world inputs still need validation. Legal and normative judgment remain institutional. Final responsibility remains human.
As machines become more autonomous, independent verifiability becomes more important because a machine cannot serve as the ultimate trust source merely through reputation or “personality.” The aim is narrower: whenever a dispute can be transformed into a factual question, rely less on repeated belief and more on evidence that can be independently checked, replayed, and reused.
If this transformation is sustained by industry data, trust can become not only a social relationship backed by institutions but also an infrastructure whose cost can be accounted for, quality compared, reuse measured, and productivity tested. The present paper establishes that accounting framework and its comparison rules; it does not claim to deliver a finished economy-wide statistic. USDC shows how the process can begin with a bounded trust product. Other industries can then build their own satellite accounts under the same principles.
References and Adjacent Literature
A. Core Academic and Peer-Reviewed Research
Coase, Ronald H. “The Nature of the Firm.” Economica, 1937.
Samuelson, Paul A. “The Pure Theory of Public Expenditure.” The Review of Economics and Statistics 36(4), 1954: 387–389. DOI: 10.2307/1925895.
Buchanan, James M. “An Economic Theory of Clubs.” Economica 32(125), 1965: 1–14. DOI: 10.2307/2552442.
Ostrom, Elinor. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press, 1990.
Williamson, Oliver E. The Economic Institutions of Capitalism. Free Press, 1985.
Jensen, Michael C., and William H. Meckling. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.” Journal of Financial Economics 3(4), 1976: 305–360.
North, Douglass C. “Institutions.” Journal of Economic Perspectives 5(1), 1991: 97–112.
Budish, Eric. “Trust at Scale: The Economic Limits of Cryptocurrencies and Blockchains.” Quarterly Journal of Economics 140(1), 2025: 1–62. DOI: 10.1093/qje/qjae033.
Lu, Saite, and Diane Coyle. “Social trust and productivity: evidence from European NUTS-2 regions.” Regional Studies 60(1), 2026, Article 2653002. DOI: 10.1080/00343404.2026.2653002.
Townsend, Robert M. “Optimal Contracts and Competitive Markets with Costly State Verification.” Journal of Economic Theory, 1979.
Diamond, Douglas W. “Financial Intermediation and Delegated Monitoring.” Review of Economic Studies, 1984.
Neu, Dean. “New Stock Issues and the Institutional Production of Trust.” Accounting, Organizations and Society, 1991.
Philippon, Thomas. “Has the US Finance Industry Become Less Efficient? On the Theory and Measurement of Financial Intermediation.” American Economic Review, 2015.
Catalini, Christian, and Joshua S. Gans. “Some Simple Economics of the Blockchain.” NBER Working Paper / Communications of the ACM.
B. Policy and Institutional Cases
European Commission. EU-wide digital Once-Only Principle for citizens and businesses, 2017.
U.S. Bureau of Labor Statistics. “Occupational employment in commercial banking, 1987–90,” Monthly Labor Review, 1993; and 2023 Occupational Employment and Wage Statistics: Credit Intermediation and Related Activities.
Conference of State Bank Supervisors (CSBS). 2025 Annual Survey of Community Banks; Thomas F. Siems and Nathan Ross, “Do Banking Regulations Disproportionately Impact Smaller Community Banks?”, 2025.
Public Company Accounting Oversight Board (PCAOB). Data Points: Registered Firm Staffing Trends, July 2025.
C. Official and Industry Technical Sources
Ethereum official materials. Ethereum’s Layer 2 Ecosystem — Shared Security / No Bootstrapping.
Circle Internet Group, Inc. Form 10-K for the year ended December 31, 2025 (filed with the U.S. SEC on March 9, 2026); 2026 Proxy Statement (filed on April 1, 2026).
Circle. Transparency & Stability / USDC monthly assurance reports.
Consensys. “Ethereum is the Trustware to Power a Digital World Economy,” 2025.
MITRE. Framework for Continuous Remote Attestation in Confidential Computing Environments, Version 0.9.6, Public Review Draft, July 16, 2026.
Mastercard. “When AI Starts Buying for You, Trust Becomes the Product / Verifiable Intent,” March 5, 2026.
D. Prior Theoretical and System-Design Work by the Author
Zhu, Weisha, and Scott Shields. From Double-Entry Bookkeeping to Verifiable Finance, 2026. [Chinese-language work; background on transparent banking, public trust roots, and verification embedded in business processes.]
Zhu, Weisha. Verifiable Financial Systems White Paper (V1.1), 2026. Working manuscript.
Chainless.hk. “Beginner Q&A on the Chainless Platform and DW20 Stablecoin (I),” July 3, 2024. [Public website material referencing the earlier WEB3.0 Chainless Financial Platform White Paper.]
E. Frontier Preprint
Rhodes, James, and George Kang. “Proof of Execution: Runtime Verification for Governed AI Agent Actions.” arXiv:2607.05397, April 26, 2026. Preprint; not peer reviewed.
