The Missing System Theory addresses harms that Coasean bargaining cannot carry because no party owns the shared system condition at stake. Public health, ecological capacity, market trust, fiscal capacity, and institutional integrity are not rights one party can sell, waive, or bargain away on behalf of the system.

Coasean bargaining works when the harm can be assigned to a right, owned by a party, and traded at tolerable cost. MST identifies the structural break when that condition fails: the system-welfare coordinate is outside the bilateral payoff vector. The System Asset Pricing Model (SAPM) then measures the mispriced system-welfare damage by comparing independently measured system harm with the revenue associated with that harm.

The Conflictoring Protocol and Decision Accounting extensions specify the game change: make system welfare measured, recorded, and decision-relevant so the institutional rules reduce the damage. The larger economic move is that the missing cost becomes visible, rankable, and actionable inside the decision architecture.

Hayek's knowledge problem supports open system-welfare measurement. SAPM does not allocate output. It reports β_W = ΔW / Π: independently measured annual system-welfare loss divided by annual industry revenue on the same boundary.

That number gives dispersed actors usable information. Investors can reprice risk, insurers can adjust coverage, courts can assess liability, regulators can demand a decision record, and firms can redesign the activity. The new information is the missing system cost: once it is measured, ranked, and tied to reform moves, market and legal actors can respond without pretending that the bilateral price already carried the whole social cost.

The governance rule is transparency: source-grounded inputs, stated boundaries, sensitivity ranges, and a record that can be challenged.

The ledger separates destroyed welfare from transfers. A shifted payment is removed. A death, poisoned aquifer, lost antibiotic effectiveness, fiscal-capacity loss, market-trust loss, or cleanup burden remains because system welfare has been reduced.

Numerical claim withheld pending source admission and independent re-estimation.

The answer to the identification objection is reproducibility: show the source, show the channel, show the denominator, show the overlap correction.

W is the system-welfare coordinate, not an aggregation of preferences. It asks whether the aquifer remains usable, whether antibiotics still work, whether the fiscal base is intact, whether the benchmark is trustworthy, or whether the land, air, or water can absorb more damage without losing function. Sen's and Arrow's impossibility theorems govern the aggregation of individual rankings.

MST measures the condition of the support system affected by the transaction. The proof claim is that W is not recoverable from agent payoffs alone. A payoff-space analysis can show what the named parties received, but it does not carry the full system-welfare coordinate unless W is independently measured.

W is the system-welfare coordinate, not a vote tally or a preference ranking. Arrow constrains rules that convert individual preferences into a collective ranking. MST asks a different question: whether the transaction raises or lowers the welfare of the system the parties depend on.

The 8-outcome taxonomy uses a binary indicator: c = 1 when the system is preserved and c = 0 when it degrades. For a renewable resource this is a physical threshold — a preservation floor W₀, the stock level below which recovery fails — but most domains have no such floor: welfare loss is unbounded below, and c = 0 marks qualitative degradation. The binary indicator is c. The underlying welfare damage can move by degrees: benchmark integrity, public-health capacity, ecological capacity, and fiscal capacity can all degrade incrementally.

That is why W is measured independently rather than derived from preferences.

Decision Accounting is a seventeen-field, reconstruction-capable governance record for materially consequential organizational decisions. It records who decided, what was decided, when, where, why, on what evidence and authority, what uncertainty remained, what prediction was made, what communication was authorized, and what system welfare impact was accepted, among other fields and record elements. The software and governance controls also require a reconsideration trigger.

It belongs in behavioral economics because it changes the choice architecture of organizational judgment: it makes the rationale and evidence salient before approval, adds friction to unsupported confidence, forces a forecast while the decision can still be revised, creates a future scoring rule through Field 16, prediction, and makes the wider system consequence visible through Field 17, system welfare impact. The behavioral premise is simple: showing the work improves the work.

The MST addition is that organizations must also show the system-welfare cost and the game-change moves available to reduce it. The data addition is that each record becomes structured judgment data: what the organization decided, why, who owned the choice, what was predicted, what later event tested the prediction, and what outcome followed. The testable claim is whether that record discipline improves reasoning quality, prediction accuracy, accountability, decision support, and learning from outcomes.

They are different impossibilities about different things. Myerson-Satterthwaite is about two parties who may fail to trade efficiently because each has private information about the deal. The problem lives inside the buyer-seller bargain.

The Missing System Theory asks a different question: what happens when the buyer and seller can both gain while the system affected by the bargain is not represented in their payoff vector. The missing piece is the system-welfare coordinate that neither trader owns, prices, or carries unless the institution requires it. That is why a mechanism can solve the private-information problem and still produce a Hollow Win.

The theorem is not refuted; its domain changes. MST is a conditional result: given the three axioms — overlapping interests, system dependence, and system independence, where W is not a function of the parties' payoff vector — the Hollow Win is the equilibrium. A theorem does not become false when its premises stop applying. The Missing System Theory is both the map of the trap and the map of the door: the same statement that locates the missing coordinate names the way out, which is to make W present.

That is what the program does. Decision Accounting, Conflictoring, and game change do not refute MST; they are responses to it. Decision Accounting records the system-welfare consequence, Conflictoring creates parties with an incentive to act on it, and game change alters the rules so private advantage comes to depend on system effects. Once W is carried in financial statements, reserves, disclosures, covenants, liabilities, and counterparty terms, the parties are no longer contracting over the old private-payoff vector alone, and the original setup no longer describes the augmented game.

So MST is permanent as a theorem but contingent as a diagnosis. W is not made a natural function of the payoffs; it is made institutionally coupled to them, and that coupling has to be maintained — through Decision Accounting, liability, disclosure, enforcement, investor discipline, counterparty terms, insurance, ratings, regulation, and Conflictoring. Let those systems decay and system independence reappears, and the Hollow Win returns with it. The program does not erase the loss once and for all; it converts a one-time impossibility into a standing maintenance problem.

The win condition, then, is not that welfare loss vanishes — some domains carry residual harm no rule can remove. It is that avoidable β_W is driven toward zero while residual β_W is named, priced, and governed. The deepest confirmation of the theorem would be a world where it rarely bites, because the institutions finally learned its lesson. The program seeds the destruction of the world in which MST is most damaging. It does not seed the destruction of the theorem.

No. In SAPM, β_W is not introduced as a stock-return factor. It is a welfare-accounting ratio: β_W = ΔW / Π, meaning annual system-welfare loss divided by annual industry revenue on the same boundary. The factor-zoo objection applies to a different claim: that a signal predicts traded asset returns after controlling for known factors.

The harder market question is why public markets do not already price system-welfare exposure. The answer is that markets price expected firm cash flows, liability, regulation, insurance cost, capital cost, reputation, and disclosed risk. They do not automatically price full ecological capacity, public-health capacity, benchmark integrity, fiscal capacity, or market-trust loss when those costs remain legally, informationally, or temporally disconnected from the security.

The return-side paper therefore asks the narrower finance question: when does the excluded welfare cost enter asset prices? The current answer is not a broad normal-period β_W premium, but a legibility-event channel: litigation, regulation, settlement, writedown, insurance repricing, financing restrictions, or disclosure can connect the system cost to expected cash flows. The broader program adds a propagation claim.

As SAPM measurement, Decision Accounting records, Conflictoring channels, litigation, regulation, insurance, and investor scrutiny make system-welfare exposure more legible, markets should become more sensitive to those events. The first expected effect is greater pricing of system-welfare legibility events, not a universal β_W factor premium.

Field 17, system welfare impact, works because it is timestamped before the outcome. It forces the decision-maker to record the expected system welfare impact, why that assessment was accepted, supporting evidence, uncertainty, and reconsideration trigger while the decision is still live. Later review can compare the prediction with the outcome.

Proxy-gaming is handled in two layers: the record fields supply who, what, when, where, why, evidence, authority, uncertainty, communication, and Field 16, prediction, among other fields; the software adds outcome scoring and anti-gaming review.

Conflictoring answers the capture problem by spreading enforcement pressure across the seven recurring lanes rather than relying on one regulator: employees and whistleblowers, CEOs and boards, regulators, policymakers, shareholders, plaintiff litigators with lawful access in the relevant setting, and communities and the affected public.

The point is to make the Hollow Win harder to preserve. Once the system cost is measured and reaches enough independent lanes through lawful channels, the firm must manage legal risk, employee escalation, board responsibility, investor pressure, policymaker attention, and regulatory exposure at the same time.

The design requirement is that enough independent lanes must receive or learn the measured system cost, have authority or incentive to act on it, and remain hard to suppress at the same time. When that threshold is met, the firm cannot preserve the Hollow Win by capturing a single regulator or neutralizing one audience; the system-welfare cost becomes part of the private decision environment.

The framework does not depend on gatekeepers being naturally public-spirited. It changes the audience problem. A normal gatekeeper can conform to management because management is the audience that pays, evaluates, and renews the relationship. Decision Accounting makes the decision record future-facing: it may be reviewed by multiple audiences with different incentives.

That changes the writer's incentives. The chief decision officer has to write for management, auditors, regulators, counsel, future managers, and litigation risk at the same time. The record has to survive later review by parties who care about different failures: rationale quality, evidence quality, authority, prediction accuracy, disclosure consistency, system-welfare impact, and weak explanation of the chosen path, among others. A record tailored to please one audience can become damaging when read by another.

Conflictoring is the reason that matters. It does not mean the record is public by default. Access remains controlled by governance rights, audit scope, examination authority, required disclosure, subpoena, court-supervised discovery, or other lawful channels. The point is multi-audience reviewability: because the writer cannot know which authorized audience will matter later, the best strategy is to meet the record-quality standard.

Numerical claim withheld pending source admission and independent re-estimation.

Goodhart risk means the metric becomes a target and people start optimizing the number instead of the underlying system. The first control is methodological: the SAPM estimate must remain tied to sources, channel boundaries, denominator rules, uncertainty, and revision history. The second control is organizational: if a firm uses the estimate in a Decision Accounting record, the record should state which SAPM source or version it relied on, what uncertainty it accepted, what prediction it made, and what reconsideration trigger would apply. Decision Accounting does not replace SAPM expertise; it records how that expertise was used in a live decision.

What is supported now is the mechanism and the experimental design, not a completed field result. Calvano-style learning supports the premise that pricing agents can reach supracompetitive outcomes without an explicit human cartel.

The interesting DA test is algorithm-centered. Give pricing agents the same market environment, but require the treatment agents to complete the seventeen Decision Accounting fields themselves, including who, what, when, where, why, authority, evidence, uncertainty, Field 16, prediction, communication logic, and Field 17, system welfare impact, among other fields. Then compare their pricing path against agents that learn prices without that record obligation.

The outcome is whether self-recording changes algorithm behavior: do the agents still converge to supracompetitive pricing, do they detect the welfare-damaging equilibrium, do they alter their policy, or do they produce a record that exposes the collusive logic after the fact? The deeper data question is whether the record gives the system a usable learning trace: the choice, rationale, prediction, price path, outcome, and welfare consequence in one structured timeline. The open empirical question is whether a reconstructable decision record changes learned pricing behavior in live or simulated pricing systems.

System-Welfare-Adjusted GDP directly answers the Deaton criticism. GDP is the kind of single-dimensional welfare statistic Deaton warns about: it records production while leaving the social cost of that production outside the headline number, or even counting repair activity as additional output. The 2009 Stiglitz-Sen-Fitoussi report made the same basic point and recommended broader well-being dashboards.

Dashboards help, but they leave GDP intact as the dominant headline number. System-Welfare-Adjusted GDP makes the missing correction inside the headline accounting itself: it keeps GDP as the production measure and adds the contra account for system-welfare cost. Mortality, cleanup, illness, enforcement, ecological degradation, fiscal burden, and future damage are priced on the same boundary as the output that generated them.

That is the Kuznets warning in operational form. The estimate is reported with uncertainty bands and sensitivity tests because the claim is comparative and reconstructable, not false precision. The policy value is that it shows which output depends on system damage, how large the damage is, how domains rank, and where reform can convert that loss into a dividend.

The move from G to G-prime is a mechanism-design claim because it changes the game form: who must record the decision, what information must be produced, who can later inspect or act on the record, and which payoffs become relevant before the Hollow Win is selected. MST identifies the missing welfare coordinate.

SAPM measures the welfare damage. Decision Accounting creates the decision record at the point where the damage is accepted. Conflictoring gives independent actors incentives to use that record when the system-welfare cost is ignored.

The mechanism is welfare completion: system welfare enters the operative decision architecture before approval, instead of appearing later as a disclosure after the private payoff structure has already controlled the decision. The practical rule is that forcing the decision-maker to show the work can improve the decision itself.

The formal statement — which strategies the record regime removes, and under what equilibrium concept the transformed game improves welfare — is in the linked paper; the site claim is the game-form change, not an equilibrium theorem.

Williamson's test is comparative governance cost: does the governance structure improve the transaction enough to justify the burden it adds? For Decision Accounting, the relevant benefit is broader than later reconstruction. The record can improve the decision before approval by forcing the decision-maker to show the work: who, what, when, where, why, authority, evidence, uncertainty, Field 16, prediction, communication, and Field 17, system welfare impact, among other fields.

It can also improve review after approval by making the decision easier to audit, score, learn from, and place inside a recoverable point-in-time governance state if the decision later fails. The record also becomes reusable organizational data: the firm can compare decisions, non-decisions, later events, predictions, outcomes, and weak-answer patterns across time. The Calvano follow-up is one way to test the premise.

If pricing agents that complete Decision Accounting records move closer to competitive pricing than agents that learn without the record, that would be evidence that record discipline changes decision quality. The result would matter to antitrust officials, companies, and owners because illegal or collusive pricing exposes the firm to enforcement, litigation, reputational damage, and governance failure, even when short-run profits rise. For algorithmic agents, the marginal cost of completing the record may be especially small.

The transaction-cost question is therefore whether the incremental value of better decision-making, better accountability, faster learning, lower legal risk, lower forensic recovery cost, and better decision-support data exceeds the incremental burden of creating the record. DA is justified for materially consequential decisions where that expected governance value is larger than the record cost; it is not justified for routine decisions where the record burden would exceed the likely improvement.

Decision Accounting would not guarantee crisis prevention, but it would make critical decisions and non-decisions visible while they can still be challenged. Silicon Valley Bank is the non-decision example: the bank repeatedly chose, in effect, not to hedge, and it left the Chief Risk Officer role without a permanent replacement for roughly eight months while interest-rate risk was rising.¹⁸

DA matters because the decision timeline can be cross-checked against the event timeline: Fed inflation warnings, Fed speeches about possible rate increases, futures-market signals, liquidity warnings, deposit-flow changes, risk-leadership vacancies, and internal risk reports. That paired timeline is the decision-data thesis in practice: decisions, non-decisions, events, triggers, predictions, outcomes, and missed reconsideration points in one recoverable sequence. The 2008 financial crisis is the product-governance example.

DA would not automatically identify a mortgage-backed security, CDO, synthetic CDO, rating model, or warehouse decision as fraudulent. It would force a contemporaneous record of who approved the product, what was approved, when, where in the governance chain, why it was approved, what representations were made to buyers, what loan-quality evidence was used, what model assumptions were accepted, what conflicts were disclosed, what adverse evidence was ignored, what prediction was made, and what system consequence was accepted.

Fraudulent or reckless financial products become easier to identify when the product record can be compared with the underlying loan files, rating assumptions, internal emails, market signals, and later performance. The deterrent is that decision-makers know management, auditors, regulators, shareholders, and possibly later litigants may read the record through authorized channels, so the best strategy is to make the record strong under the record-quality standard.

The software role is important: it should prompt the user to consider missing non-decisions when new evidence makes an unrecorded choice material. The claim is disciplined decision-making plus recoverable point-in-time governance states, not hindsight certainty.

Conflictoring classifies enforcement powers, not organization types. Ostrom's communities and user associations matter when they hold one of the powers needed to act on system harm. The six recurring agents are functional.

Insiders and whistleblowers can reveal information the firm would otherwise suppress. Boards and executives can change the decision internally. Investors and capital allocators can change financing, valuation, voting, or compensation pressure.

Regulators can examine, sanction, condition approval, or require remediation. Litigants can use lawful process to turn hidden harm into liability. Policymakers can change the rule architecture when private and regulatory channels are insufficient, including through international bodies where the domain requires cross-border coordination.

Ostrom's polycentric governance is therefore not excluded; it appears when communities, users, cooperatives, local monitors, or affected parties occupy one of those lanes with real observation and cost-imposition power. The question is functional: who can observe the system harm, impose a cost on the Hollow Win, and keep the enforcement channel independent?

β_W adds attribution and scale. Stiglitz and Greenwald explain why imperfect information and externalities make markets inefficient; SAPM asks how large the system-welfare loss is, which activity produced it, and how to compare that loss across domains. The inputs are not invented inside the model.

SAPM uses public evidence and expert estimates from existing literatures: mortality, illness, cleanup, ecological damage, productivity loss, fiscal burden, financial fragility, or other domain-specific harm channels. The new contribution is the MST/SAPM structure: identify the system-welfare loss that the bilateral transaction does not carry, match the harm channels to the revenue boundary that produced them, correct overlap, report uncertainty, and express the result as β_W = ΔW / Π. Externality theory says market failure exists.

β_W makes the failure attributable, comparable, and usable for governance.

Most SAPM domains are repeated, observable, high-frequency harms with uncertain parameters, not black-swan forecasts: alcohol illness and mortality, gun violence, opioid addiction, plastics in drinking water, air pollution, tobacco disease, gambling losses, payday lending costs, or tax-base erosion. The question is not whether an unknowable catastrophe will occur.

Numerical claim withheld pending source admission and independent re-estimation.

Taleb's warning still matters for domains with true tail risk, so SAPM reports sensitivity under different distributions and does not rest the theorem on a precise point estimate. But for much of the 61-domain panel, the model is measuring common recurring damage, not pretending to predict a once-in-a-century event.

Decision Accounting is built for bounded rationality. It asks for a contemporaneous account of the decision basis: who decided, what was decided, when, where, why, what was known, what evidence was used, what uncertainty remained, and what prediction was made, among other fields. Satisficing can be a valid recorded reason: three options, limited evidence, a deadline, and a chosen path.

The record disciplines memory and later accountability by preserving the reasoning before outcome knowledge rewrites it. Over time, those bounded judgments become institutional memory. The organization gets a structured record of judgment: the constraint faced, the choice made, the forecast attached to it, the event that later tested it, and the result.

That is the data needed for training, decision support, and calibration.

Numerical claim withheld pending source admission and independent re-estimation.

The proven-model requirement disciplines the claim. Before calling a domain institutionally solvable, Policy Lab asks for a country, program, or rule architecture that has already produced measured improvement. Path dependence means the old system has defenders, habits, contracts, sunk investments, and administrative routines that make reform difficult.

A proven model does not make transfer easy; it shows that the remaining barrier is institutional design and political implementation rather than an irreducible welfare floor.

Olson's politics is the reason Conflictoring uses enforcement lanes. The framework does not assume diffuse beneficiaries will organize on their own. It asks which actors already have concentrated incentives or duties to act when system-welfare harm is made measurable: employees and whistleblowers, CEOs and boards, regulators, policymakers, shareholders, plaintiff litigators with lawful access in the relevant setting, and communities and the affected public. Auditors, courts, and journalists may matter in particular cases, but they operate through or amplify those lanes rather than expanding the canonical six.

Those actors do not all see every decision record. The point is controlled, legally appropriate review by multiple audiences with different incentives. The design succeeds when enough independent lanes can make the system-welfare cost privately relevant, so the Hollow Win is no longer the rational choice.

Decision Accounting adds the missing evidence layer for optimal regulation under asymmetric information. Laffont-Tirole asks how a regulator can design incentives when the firm knows more about its costs, actions, and risks than the regulator does. DA addresses the prior evidentiary problem: the regulator often lacks a contemporaneous record of the decision that created the risk.

The record shows who acted, what was decided, when, where, why the firm acted, what evidence it relied on, what Field 16, prediction, it made, what communication position it authorized, and what Field 17, system welfare impact, it accepted, among other fields. That matters for both hidden information and hidden action. It gives regulators, auditors, boards, and other authorized reviewers a decision record to inspect before the firm can convert uncertainty into an after-the-fact story.

Across many decisions, accepted risks, failed predictions, ignored triggers, later events, and improved controls become comparable supervisory evidence. Conflictoring then defines which actors can use that record, through which lawful access path, and with what incentive to act. DA does not replace optimal regulation; it supplies the decision-level evidence that makes better incentive design possible.

System welfare does not replace distributional analysis; it gives distributional analysis a second line to track. SAPM first measures whether the shared support system is being depleted: the tax base, public-health baseline, ecological capacity, institutional trust, or financial-system integrity. Distributional analysis then asks who gained from that depletion, who avoided the cost, and who was left carrying it.

Tax havens show the distinction. The system-welfare loss is the erosion of fiscal capacity: governments lose revenue needed to maintain public goods, enforcement, infrastructure, health, and social insurance. The distributional harm is that mobile capital and high-wealth taxpayers can shift income or assets offshore while less mobile workers, consumers, and domestic firms bear more of the tax burden or receive weaker public services.

β_W estimates the size of the system loss. Distributional analysis identifies the incidence of that loss. Together they show both how much of the shared system was depleted and who benefited from the depletion.

Limited arbitrage begins with a traded price that may be wrong. A smart investor can see the mispricing, but capital limits, timing risk, career risk, or coordinated noise trading may prevent the correction. β_W addresses a different failure.

Many system harms have no traded price to correct in the first place: public-health depletion, fiscal erosion, ecological damage, institutional trust, or benchmark integrity. No arbitrageur owns a claim that pays off when those systems are preserved, so the market can remain blind even when the harm is real. β_W creates the missing signal by asking how much system welfare is being lost per dollar of revenue in the activity that produces the harm.

That gives boards, investors, regulators, courts, employees, and policymakers a concrete exposure to act on. The metric is useful because it turns an unpriced system loss into a visible governance and market-risk signal. You cannot manage what you do not measure.

The Postnieks program shares Raworth's and Mazzucato's premise: public value has to be designed, measured, and protected. Raworth's Doughnut Economics asks whether economic activity stays inside a safe social and ecological operating space. Mazzucato's mission economics asks whether the state, firms, and finance are organized around public-purpose missions rather than passive repair after failure.

β_W supplies the missing contra account for both projects: which activities are draining public health, ecological capacity, fiscal capacity, institutional trust, or financial stability, and by how much per dollar of revenue. MST explains why those costs are missing from the current game. SAPM measures the depletion.

System-Welfare-Adjusted GDP shows the macroeconomic scale. Decision Accounting records where the depletion is accepted in live organizational decisions. Conflictoring identifies the lawful review and enforcement channels that can make the cost matter.

Policy Lab and Reform Pathfinder then ask which game-change moves reduce the loss. The point is to give mission-oriented redesign a map of what must be reduced, replaced, governed, or prohibited.

Field 17, system welfare impact, creates disclosure and litigation exposure. That is an undisputed part of the design.

The answer is legal architecture. The Chief Decision Officer should sit inside, or directly under, the legal function; the recommended protocol is a deputy-chief-legal-officer model because system-welfare records, securities disclosure, privilege, and litigation risk have to be coordinated. Coffee's gatekeeper problem is central: lawyers, auditors, ratings intermediaries, and disclosure professionals need a real record to test.

The Delaware oversight case In re Caremark International Inc. Derivative Litigation points in the same direction: directors must make a good-faith effort to maintain reasonable information and reporting systems for material risks. Field 17, system welfare impact, makes the accepted system consequence part of that board-level governance record. A careless Field 17, system welfare impact, can become an admission.

A disciplined Field 17, system welfare impact, can become process evidence: the firm identified the issue, explained why the choice was made, assigned authority, stated uncertainty, made Field 16, prediction, set a reconsideration trigger, and aligned external communication with the internal record, among other fields and record elements. Caremark makes record quality legally important; it does not make any record an automatic defense. The shareholder-value answer is that better governance should lower risk.

The Decision Accounting papers argue that high-quality governance records can reduce cost of capital, potentially on the order of up to 200 basis points in the modeled estimates. That upside is why the exposure can be worth taking when the system is executed carefully, under counsel supervision, with disclosure controls that make the decision record and public record consistent.

LIBOR teaches that benchmark manipulation is fixable when regulators treat the benchmark as public infrastructure rather than a private market convention. The scandal was solved by changing the benchmark architecture, moving away from bank submissions, and forcing markets toward more reliable reference rates. FX fixing needs the same governance lesson.

The fix includes governance before the chat-room scandal: who approved the fixing method, what method was approved, when, where in the governance chain, why, what manipulation risk was accepted, what volume concentration occurs around the window, what controls monitor order handling, and what event would trigger redesign. Decision Accounting moves those questions to the decision point. The benchmark has private users, but its failure creates public-system harm: distorted contracts, damaged price discovery, lost trust, and institutional legitimacy costs.

The supported claim is that FX fixing should be governed like critical market infrastructure, and LIBOR shows that benchmark redesign is possible when regulators force the issue.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Where the work departs from federal defaults, mainly on discounting harms to future generations, the departure is stated as a modeling choice with a reason and sensitivity tests.

It is a real sensitivity, and the work should show it explicitly. The current model uses one stated value of statistical life so that every domain is measured on one common welfare ledger. That choice avoids valuing identical deaths differently because they occur in countries with different incomes, and it departs from standard benefit-cost practice that sometimes transfers VSL by income elasticity.

The defensible presentation is two-layered: report the common-ledger estimate as the main comparability case, and sensitivity-test income-adjusted VSL transfer for globally scoped domains such as human trafficking, child labor, illicit drugs, and pollution-heavy supply chains. The claim that should survive is sign and rank: whether the domain remains welfare-destructive and how it ranks under alternative VSL transfer rules. If a domain changes rank materially under income-adjusted VSL, the table should show that rather than hide it.

Specific answer: counted against the canonical seventeen DA fields, DORA directly overlaps with about 13 of 17: who, what, when, where, why, evidence, authority, training, review, stakeholders, consequences, constraints, and communication. The overlap comes from governance accountability, management-body responsibility, technology-risk documentation, asset and dependency mapping, training, incident recording, root-cause analysis, reporting, third-party technology-risk controls, and contractual exit requirements.

DORA partly overlaps with uncertainty and Field 16, prediction, through risk assumptions, incident scenarios, expected recovery, testing, and concentration-risk analysis. DORA does not require Field 15, alternatives, and does not require full Field 17, system welfare impact. It asks for narrower system-impact facts: technology resilience, critical functions, dependencies, concentration risk, client impact, economic impact, and recovery.

The EU AI Act directly overlaps with about 15 of 17: who, what, when, where, why, evidence, authority, training, review, stakeholders, consequences, constraints, uncertainty, communication, and Field 16, prediction. It partly overlaps with Field 17, system welfare impact, through health, safety, fundamental-rights, affected-person, and systemic-risk concepts. It does not require the full Field 15 alternatives account or the full DA welfare-impact account.

The decision-data point is that DORA and the EU AI Act create many required records, but those records can remain fragmented: risk assessments, logs, human-oversight notes, incident reports, testing records, recovery decisions, and post-market monitoring. DA supplies the common decision record that links those materials to the decision that used them, the prediction made, the event that later tested them, and the outcome that followed.

It adds the decision record beneath the supervisory file. The model-risk guidance line that began with SR 11-7, and now runs through SR 26-2 and OCC Bulletin 2026-13, covers model development, validation, use, limitations, governance, controls, and review. BCBS 239 governs risk-data aggregation and reporting.

Decision Accounting records the choices those regimes depend on: who approved the model or data architecture, what was approved, when, where in the governance process, why it was approved, under what authority, on what evidence, what uncertainty remained, what Field 16, prediction, was made, and what Field 17, system welfare impact, was accepted, among other fields. Supervisors still apply model-risk guidance and BCBS 239. DA gives them the dated record of the decision they would otherwise have to reconstruct after the fact.

Across models and risk-data systems, model choices, data-governance choices, validation warnings, overrides, later events, predictions, and outcomes can be reviewed as a sequence.

It would have forced records for the upstream decisions that made the monitoring failure possible. The TD Bank enforcement record points to three decision points: keeping anti-money-laundering resources inside a flat-cost budget while the bank and its risks grew; leaving major transaction channels outside automated monitoring; and launching or operating products and services without controls matched to their money-laundering risk.

Numerical claim withheld pending source admission and independent re-estimation.

Decision Accounting would not wait for a suspicious-activity report. It would require a record of who approved the budget constraint or product channel, what was approved, when, where in the governance chain, why those choices were made, what uncertainty remained, what Field 16, prediction, was made about money-laundering exposure, and what Field 17, system welfare impact, was accepted for the financial system, among other fields.

Senior management, board members, auditors, and regulators could then have read those decision records, cross-checked them against growth, backlog, product, and enforcement-event records, identified the accumulating risk, and managed it before it became a criminal, regulatory, and financial liability.

Specific answer: counted against the canonical seventeen DA fields, the UK Senior Managers and Certification Regime directly overlaps with about 8 of 17: who, what, when, where, authority, training, review, and communication. It does this through named senior managers, statements of responsibilities, responsibility maps, certification, conduct-rule training, governance documentation, and regulatory accountability for reasonable steps.

It partly overlaps with evidence, constraints, and consequences because a senior manager may need to show what reasonable steps were taken, what limits or reporting lines shaped the role, and what happened inside the controlled area. It does not require the full DA fields for why, uncertainty, Field 16, prediction, Field 15, alternatives, or Field 17, system welfare impact. The regime supplies a legal accountability setting.

Decision Accounting supplies the complete reconstructable decision record that makes that accountability easier to test.

It has to be legally engineered. Careless welfare-impact records create admissions risk. Missing records create reconstruction risk: after a failure, the firm still has to explain who knew what, when, under what authority, and why the choice was made.

Decision Accounting addresses that tradeoff by separating legal advice from governance evidence and by making the record show disciplined process: why the choice was made, what evidence was used, what uncertainty was accepted, what Field 16, prediction, was made, what residual risk was accepted, what communication position was authorized, and what review triggers were set, among other fields and record elements. The Delaware oversight case In re Caremark International Inc. Derivative Litigation is the right legal reference point.³⁷

Caremark stands for the proposition that directors must make a good-faith effort to maintain reasonable information and reporting systems for material risks. A serious decision record can help the corporation show that the board and management had a process for identifying risk, assigning authority, reviewing evidence, and responding before harm matured. That is useful litigation evidence and useful governance evidence.

The shareholder-value point is the same: a firm with disciplined records should be easier to supervise, easier to defend, and less exposed to surprise governance failures. The practical rule is counsel-supervised design, with the Chief Decision Officer inside or directly supervised by the legal function where disclosure risk is material.

System-Welfare-Adjusted GDP should start as a cited companion statistic and then, if the work survives review, become a candidate contra account paired with GDP inside the national-accounts conversation. That sequencing matters because the System of National Accounts is a formal statistical standard, updated most recently through the 2025 SNA process, and a new welfare-loss ledger needs source review, boundary testing, overlap correction, and uncertainty discipline before official adoption.

The substantive claim is that GDP remains the dominant headline number, and dashboard proposals have not solved the weakness identified by Kuznets, Deaton, and the Stiglitz-Sen-Fitoussi report: GDP records production without carrying the system-welfare loss attached to some of that production. Dashboards are still useful for distribution, health, environment, sustainability, subjective well-being, and other dimensions that no single number should carry.

System-Welfare-Adjusted GDP changes the GDP frame itself by recording annual system-welfare loss from damage, cleanup, illness, enforcement, ecological degradation, fiscal burden, financial fragility, and future costs. It also gives governments a ranked reform map: which sectors are producing apparent output by consuming system capacity, and where the largest reform dividends are likely to be.

Numerical claim withheld pending source admission and independent re-estimation.

Policy Lab uses the proven country case as a translation template. It asks: what decision kept the Hollow Win in place, which local actor has authority, which Conflictoring lane can act at the lowest total prevention cost, which coalition blocks the change, what administrative capacity is missing, what sequencing is required, and what evidence would show the local reform failing. The country case proves that a game change has worked somewhere. The local record still has to show that the same kind of game change can work here.

Capture is the design problem Conflictoring is built around. Stigler's warning is that a regulator can be influenced, delayed, starved of information, or politically constrained. Conflictoring does not rely on the regulator alone.

It focuses on seven agents: employees and whistleblowers, the CEO and board, regulators, policymakers, shareholders, plaintiff litigators with lawful access in the relevant setting, and communities and the affected public. The regulator is one lane, not the whole enforcement architecture. The game-change protocol asks whether k-star has been reached: the minimum number of independent lanes needed to make the old Hollow Win more expensive than changing the decision.

Those agents do not need to coordinate or share the same motive. They can act independently from ordinary self-interest: employees protect themselves, boards reduce governance risk, regulators enforce rules, shareholders protect value, plaintiff lawyers pursue liability, and policymakers respond to public failure. They also do not all need the same documents, and the record is not public by default.

Conflictoring succeeds when enough independent lanes can change the payoff of the old game, improving system welfare for the firm in question and for the industry practice that produced the Hollow Win.

The instruments are not being replaced; they are the tools, and the framework says so — the unifying result repositions Pigou, Coase, and Ostrom as three operations on the missing coordinate without erasing the traditions, and Calabresi's cheapest-cost-avoider logic as the rule for who should bear it, which Conflictoring uses. The objection is right that these are the mechanisms. What it misses is that each carries a precondition a captured Hollow Win breaks. Pigou needs an authority that knows the external cost and is not captured; in a Hollow Win the cost is unmeasured and the regulator's own revenue often rides on the harm. Coase needs the affected party at a low-friction bargaining table, but the system has no seat — bargaining internalizes cost among the parties present and cannot reach a party that is not there. Ostrom needs an organizable local community, which diffuse, intergenerational, and global harms do not have. Capture and the missing seat are the common failure modes that neutralize any single instrument. Conflictoring is not a fifth mechanism competing with these four. It is the deployment layer that answers a question the classical toolkit does not pose: how to get the tools activated when the game is captured. Its content is the reachability result — because capture can block any single channel, the change requires the minimum number of independent reform lanes moving together, k^∗, and the lanes are simply who wields the classic tools: the regulator and policymaker carry Pigou, plaintiff attorneys carry tort and liability, communities carry Ostrom, and shareholders and whistleblowers add independent pressure. Then Conflictoring applies Calabresi's own logic as its ranking rule: across those lanes it asks which actor can most cheaply prevent, reduce, expose, insure, regulate, litigate, or reprice the harm, and deploys the cheapest set that reaches k^∗. k^∗ says how many lanes; Calabresi says which ones. Each classical instrument runs through a single gatekeeper — a regulator, a bargaining table, a community — that capture can neutralize, and the tool has no second route; Conflictoring is the theory of what to do then, with Calabresi's cheapest-cost-avoider as the selection rule.

No. SOX is aimed mainly at financial-reporting accountability and internal control over financial reporting. COSO supplies the internal-control architecture: control environment, risk assessment, control activities, information and communication, and monitoring. ISO 31000 supplies a risk-management process: scope and context, risk criteria, assessment, treatment, monitoring, recording, and reporting.

Those regimes overlap with DA on authority, evidence, review, communication, consequences, constraints, and accountability. They partly cover why, uncertainty, and reconsideration. They generally do not require Field 16, prediction, or Field 17, system welfare impact, as DA defines them.

Decision Accounting adds the full reconstructable decision record: who made the consequential choice, what was decided, when, where, why, under what authority, on what evidence, what uncertainty was accepted, what prediction was made, what communication position was authorized, and what system welfare impact was accepted, among other fields. It also adds a technical record architecture: timestamping, tamper-evident integrity through hash-chain or Merkle-style anchoring, outcome scoring, weak-answer controls, and anti-gaming review.

The business value is reusable decision data. Reconstructable decision records are organizational intellectual property because decisions, non-decisions, external events, predictions, outcomes, and welfare consequences can be studied together. That lowers forensic reconstruction cost, reduces avoidable consulting and litigation expense, improves board and regulator review, and gives management a real decision-support asset.

Better records can improve decisions, reduce Hollow Wins, preserve future earnings capacity, improve system welfare, and create shareholder value.

The theorem is substantive because it makes a recoverability claim: if two decisions have the same payoffs for the named parties but different effects on the shared system, system welfare cannot be recovered from the bilateral payoff vector alone. Example: two cement plants have the same owner profit, worker wages, customer prices, tax payments, emissions controls, and production volume. One plant is upwind of a dense neighborhood, schools, and a hospital.

The other is in a remote industrial zone with the same emissions but far fewer people exposed. The private payoff vector can be identical in both cases. The system-welfare consequence is different because the health damage depends on population exposure, which is outside the owner-worker-customer payoff vector.

A rule reading only the named parties' payoffs cannot distinguish the two cases. That is the missing-coordinate result. A PST game — a game with private-systemic tension — is a domain where private parties can gain, their action affects system welfare, and the system consequence is not recoverable from their payoffs alone.

A non-PST game lacks one of those conditions: no mutual private gain, no real system stake, or the full system consequence already captured in the payoffs (complete markets, or forms that give residual control to the affected party). Non-PST games are the exception, not the rule. Private attributes like quality, cost, and customer value are carried inside the payoff structure, but system welfare, by System Independence, is not carried fully: reputation, liability, and insurance price only pieces of it, so the residual remains. The Postnieks program is aimed at the high-damage PST domains: pollution, public-health depletion, benchmark manipulation, tax-base erosion, financial fragility, addiction markets, weapons spillovers, and similar cases where the system cost is outside the transaction.

The goal is to convert PST games into non-PST games by measuring the system cost and making it decision-relevant through pricing, liability, disclosure, governance, insurance, capital markets, regulation, or litigation. That world is still a market economy with private firms. It is a market economy in which fewer profits depend on unmeasured system damage.

β_W uses revenue because the numerator and denominator must share the same domain, time period, and activity boundary. The numerator is annual system-welfare loss, ΔW. The denominator is annual industry revenue, Π.

Profit is unsuitable because accounting choices, financing structure, tax strategy, and depreciation can move it away from the activity scale. Revenue is the auditable activity denominator. The hard part is not the division.

The hard part is making ΔW legitimate: naming the welfare channels, bounding the industry, using public sources, stripping transfers, correcting overlap, reporting uncertainty, and matching the harm to the revenue associated with that harm. Economists had many ingredients before SAPM: externality estimates, social-cost estimates, health burdens, cleanup costs, fiscal losses, environmental accounts, and cost-benefit methods.

The missing move was to organize those pieces into a repeatable industry-level ratio: annual system-welfare loss per dollar of revenue. MST supplies the reason for the numerator: the system-welfare coordinate is missing from the payoff space. SAPM supplies the measurement rule: match ΔW to Π on the same boundary, every time.

The special-case claim is formal, not a claim that public health, ecological capacity, fiscal capacity, institutional trust, benchmark integrity, or antibiotic effectiveness have traded market claims. The asset-pricing paper defines SAPM as a two-factor pricing equation: expected return depends on ordinary market risk and on system-welfare risk. Written schematically, the SAPM term is market beta times the market price of risk plus welfare beta times the system-welfare price.

When the system-welfare price is set to zero, λW = 0, the welfare term disappears and the equation reduces to the Sharpe-Lintner CAPM. That is the sense in which CAPM is a special case of SAPM: CAPM is the zero-welfare-price restriction of the larger model. This does not mean markets can never price system-welfare exposure.

Litigation, regulation, insurance repricing, disclosure, financing limits, or reputational events can make system-welfare exposure enter expected cash flows and market prices. The SAPM claim is that those channels are partial and event-dependent unless the system-welfare factor is explicitly carried in the pricing model. This is separate from the policy β_W used elsewhere in the SAPM corpus, β_W = ΔW / Π, which measures annual system-welfare loss divided by annual industry revenue.

The finance beta asks how returns load on a welfare factor; the policy beta asks how much welfare loss an activity produces per dollar of revenue. The paper connects them conceptually while keeping the two measures distinct.

A very high β_W is a diagnostic signal, and firearms show why it can be real. The numerator is annual system-welfare loss from the firearm activity being measured: mortality, nonfatal injury, emergency care, long-term disability, policing, legal-system costs, lost productivity, trauma, insurance burden, and other cited public costs, with overlap corrections so the same harm is counted once. The denominator is the revenue boundary for that same activity.

Numerical claim withheld pending source admission and independent re-estimation.

A β_W = pending re-estimation is the break-even line: one dollar of system loss for one dollar of revenue. Getting from 21.98 down to 1.0 would require eliminating about 95.4% of the measured welfare loss.

The methodological answer is therefore reproducibility: show the numerator, show the denominator, show the boundary, show the interval, and show the conservative lower bound.

The convergence statistic means that many independent governance regimes keep asking for pieces of the same decision record. It is evidence of institutional demand for the Decision Accounting architecture; it is not a claim that DA is already law or that every regime contains every field. The full DA record has seventeen fields: who, what, when, where, why, evidence, authority, training, review, stakeholders, consequences, constraints, uncertainty, communication, Field 16, prediction, and Field 17, system welfare impact.

Existing regimes repeatedly reproduce subsets of that list because consequential decisions keep requiring the same basic answers: who acted, what was decided, when, where, why, under what authority, on what evidence, with what uncertainty, with what expected result, and with what consequence for affected people and systems. The current study is author-coded working-paper evidence: it maps sixteen regimes across four continents and four legal traditions to the seventeen DA fields and reports a convergence pattern.

That supports the claim that DA did not invent arbitrary fields; it systematizes documentation demands already visible across governance regimes. It is not yet independently replicated evidence. The next validation step is to publish the regime-by-field coding table, have outside reviewers code the same source texts against the same field definitions, and then calculate inter-coder agreement.

Krippendorff's alpha would be the right statistic for that later validation because it measures how much coders agree after adjusting for chance agreement.⁴⁴ Until that replication is done, the honest claim is convergence evidence from the author's coding, not peer-reviewed proof that every field match is beyond dispute.

Decision Accounting addresses Prat's conformism problem when it is implemented as multi-audience review architecture. Prat's problem arises when the agent performs for a known audience. DA changes that incentive by making the decision record reviewable by several lawful future audiences with different powers: auditors, regulators, boards, compliance officers, legal reviewers, employee reviewers, future managers, and outside reviewers with defined access.

The control is reconstructability under independent review, supported by dated, time-stamped, tamper-evident decision records that include Field 16, prediction, Field 17, system welfare impact, and anti-gaming checks, among other fields and controls. The decision-maker's best strategy is to make the record strong under the record-quality standard because no single audience can be safely optimized for at the expense of the others.

The enterprise-learning-loop point strengthens that incentive: the record becomes part of a searchable history of judgment, events, predictions, and outcomes, so a weak record can be compared with similar decisions across the firm. That is the hypothesis: DA may solve Prat's 2005 conformism dilemma, but field data is still needed to test whether the architecture changes behavior in real organizations.

Disclosure Futility says information alone fails when the payoff structure still rewards the Hollow Win. Decision Accounting changes behavior because the record is created before approval, while the decision can still be revised. A material decision must carry a contemporaneous decision basis: who, what, when, where, why, named authority, evidence, uncertainty, Field 16, prediction, communication position, and Field 17, system welfare impact, among other fields.

The first review audience is inside the lawful governance chain: management, counsel, compliance, audit, the board, or supervisors as the setting requires. External Conflictoring agents do not all receive decision records. Regulators may see them through examination; plaintiffs may see specific records through lawful discovery; shareholders and policymakers often act through public disclosures, enforcement records, audit findings, or recurring patterns rather than direct record access.

The behavioral effect comes from foreseeable review by several independent lanes with different powers, point-in-time governance-state recovery, prediction grading, anti-gaming review, system-welfare scrutiny, and later comparison of the record with events and outcomes.

The decision-maker cannot assume the record will be judged by one friendly audience, so the best strategy is to make the record strong under the record-quality standard before the choice hardens.

The claim is not that every bad system can automatically be fixed. The claim is narrower: if the problem is caused by rules, incentives, contracts, disclosure design, enforcement design, or another institutional choice, then the repair must also change that game. A proposed fix becomes credible when it can name four things: the measurable signal of harm, the rule line that separates acceptable from unacceptable conduct, the actor with power to enforce that rule, and the reason compliance becomes the better private strategy.

That is what the Game-Change Characterization adds. It does not prove that reform will be adopted, fast, politically easy, or fully restorative. It gives Policy Lab the test: show the harm, show the rule, show the enforcer, and show why the new game works.

The three bins are impossibility, intractability, and control group. The yardstick is the constraint after the strongest realistic institutional move. Impossibility means the domain starts with a hard physical, chemical, biological, thermodynamic, or informational relationship that institutions can mitigate but cannot repeal.

PFAS is a chemical example: the carbon-fluorine bond explains the persistence floor. Liability, phase-outs, substitution, water rules, and cleanup can reduce PFAS damage, but they do not undo the persistence of released PFAS on ordinary policy horizons. Cement is another example: clinker chemistry creates process emissions, and procurement, carbon pricing, substitution, demand reduction, and production improvements can reduce cement damage, but they do not repeal the chemistry of calcination.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

It is not circular because MST does not measure W. MST says that W is not recoverable from the parties' payoff vector. That is the diagnostic claim.

SAPM then measures W separately, using cited harm channels and a revenue denominator on the same boundary. In other words, the framework first asks whether the payoff record contains the system-welfare coordinate; if it does not, SAPM supplies an independent measurement of the missing coordinate. The diagnosis is exclusion.

The measurement is ΔW / Π. Those are separate steps, so W is not being used to prove itself. The broader program then uses the measured W loss to rank domains, build System-Welfare-Adjusted GDP, and identify the Policy Lab or Reform Pathfinder move that can reduce the loss.

The seventeen-field record avoids ritual compliance through software enforcement before approval and post-decision scoring, anti-gaming review, and management feedback after the decision. A ritual works when it is cheap to fake. Decision Accounting raises that cost at the point of entry: if the user does not answer a required field properly, the software should reject the answer, explain what is missing, and give examples of an acceptable response.

The record cannot advance just because a box contains words. Evidence must cite or attach support. Authority must identify the source of authority.

Field 16, prediction, must be scoreable. Field 17, system welfare impact, must identify the affected system, boundary, expected effect, and uncertainty, among other fields. After approval, dated records, evidence attachments, authority trails, tamper-evident integrity, outcome scoring, and anti-gaming review make record quality observable.

Field 16 is later compared with the outcome, so repeated poor predictions flag a weakness in the Chief Decision Officer's judgment or calibration. Anti-gaming review flags formulaic, backfilled, or template-matched records. Those patterns become usable management data: which teams produce weak records, which decisions require rework, which predictions fail, and which interventions improve record quality over time.

Supervisors can then see the pattern and require improvement.

A decision record cannot detect every normalized shortcut by itself. Its value is that it makes drift explicit before local culture rewrites the shortcut as ordinary practice. In a Challenger-style O-ring decision, the record would force the decision chain to state who accepted the lowered standard, what standard was accepted, when, where in the governance chain, why, the evidence for accepting it, the accepted uncertainty, Field 16, prediction, the reconsideration trigger, and Field 17, system welfare impact, among other fields.

That prediction matters: the named decision-makers would have to write what they expected to happen after accepting the weaker standard. The same logic applies to a Boeing-style redundancy decision. If an engineer, manager, or Chief Decision Officer signs off on saving money by accepting less redundancy, that person's name is tied to the evidence, prediction, and accepted risk.

Showing that work can change the work because a careless prediction becomes visible to future reviewers. The enterprise-learning-loop value is cumulative: repeated exceptions, weak evidence, formulaic wording, missed triggers, later events, and adverse outcomes become a timeline that can reveal drift earlier than a single incident review.

To the extent that drift appears in formulaic wording, repeated exceptions, weak evidence, backfilled reasoning, or template-matched records, AI-enforced anti-gaming review may flag the pattern early enough for DA to prevent drift before it becomes normalized.

Field 16, prediction, does not require clairvoyance. It requires the named decision chain to write what it expects, with confidence and uncertainty, while the decision is still live. In a genuinely emergent failure, the record can say: this is the outcome we predicted, this is the probability or confidence we assigned, and this is the failure mode we did not anticipate.

That is still accountability because later review can compare the prediction, evidence, warnings, and accepted uncertainty, among other record elements, against the governance state that existed at the time. In DA, reconstruction means the software can recover that governance state in seconds: the rules, decision records, authority, and prior decisions governing the organization at a point in history.

Perrow's point sharpens the standard: complex systems make perfect foresight impossible, so DA grades the quality of reasoning under uncertainty and the prediction made before the outcome was known. The learning value comes from the timeline: decisions, event records, warnings, predictions, outcomes, and missed reconsideration triggers can be studied together.

The larger program point is that showing the work can improve the work: before accepting a risk in a complex system, the record forces the organization to identify the expected system-welfare cost and the game change considered to reduce it.

Decision Accounting can answer Scott's concern if the record preserves local judgment instead of becoming boilerplate category work. The record must show who decided, what was decided, when, where, why the choice was made, evidence used, constraints faced, uncertainty accepted, Field 16, prediction, and Field 17, system welfare impact, among other fields.

Boilerplate is a defect because it is weak decision reasoning, and DA software should flag it through weak-answer and anti-gaming controls: repeated language, template-matched answers, thin evidence, generic uncertainty, missing local facts, and weak rationale for the chosen path. Scott's objection is about legibility: standardized records can make local knowledge easier for a central authority to see by stripping away the details that made the judgment useful.

DA's seventeen fields, Brier-score feedback, anti-gaming controls, and software gates for weak answers are design choices aimed at that high-modernist legibility concern. Good decision records still require a competent Chief Decision Officer, good information from employees, serious effort by the people contributing to the record, and a genuine attempt to make the record accurate, complete, and useful for better decisions.

Properly implemented Decision Accounting reduces Burt structural-hole power. The record prevents one actor from controlling both what gets recorded and who can later read it by making the record attributable, contemporaneous, tamper-evident, and reviewable by defined audiences with different powers: boards, auditors, regulators, legal reviewers, employee reviewers, and future managers.

It preserves the reasoning that previously lived inside one privileged channel and makes the point-in-time governance state recoverable by authorized reviewers who do not depend on the original broker. The decision-data design also weakens brokerage power because the firm's knowledge is stored as structured judgment data: decisions, event records, non-decisions, prediction scores, and outcomes can be searched directly instead of filtered through one actor's summary.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

The prevention mechanism is governed scope and governed access. Decision Accounting should record materially consequential organizational decisions rather than ordinary employee behavior, and access should be limited by role, legal purpose, retention rules, and company-specific privacy policy. The current canon gives the architecture: consequential-decision scope, role-based access, tamper-evident records, and privacy as a governance-design constraint.

The detailed rules for access, retention, export, derived datasets, worker protection, and personal-exposure risk need to be written by each adopter. The Chief Decision Officer, with legal training or close legal supervision, should translate the architecture into an operating manual that fits the adopter's culture, precedent, employee handbook, existing policies, employment-law setting, and governance practice.

Decision Accounting intentionally changes behavior: it asks decision-makers to show their work, resist weak or gamed records, think through the choice, make better predictions, and write for future review by multiple audiences. That is governance discipline. The Zuboff risk is a record system used to monitor ordinary behavior or manipulate employees.

The company-level goal is better decisions. The decision-data system should be built from structured records of consequential judgment, not ordinary employee activity. That system supports decision support, prediction scoring, weak-answer improvement, risk management, cost-of-capital reduction, and shareholder value while keeping surveillance risk inside a governed-access framework.

Numerical claim withheld pending source admission and independent re-estimation.

The goal is for people to internalize MST, use SAPM to see the measured welfare loss, follow Reform Pathfinder and Policy Lab to identify game-change moves, and use the Postnieks curriculum to learn the new framework. Publication of the number is meant to change the game by making the system-welfare damage visible enough to reduce it.

No, if the calculation is done the SAPM way. SAPM does not hide the value choices inside arithmetic. It states the harm channels, sources, shadow prices, domain boundary, revenue denominator, overlap correction, and uncertainty treatment.

Numerical claim withheld pending source admission and independent re-estimation.

Powerful firms can turn any governance tool into a shield if they control the drafting, storage, and audience. Decision Accounting answers through record governance: attribution, evidence attachments, authority trails, tamper-evident integrity, privilege bifurcation, conditional safe-harbor design, and access for independent reviewers with different powers. A record drafted for friendly insiders can become a shield.

A record that must survive legal, regulatory, board, audit, and future-management review is harder to code as private protection when it has to be strong under the record-quality standard. The shield is weaker still because some records may later be tested by plaintiff attorneys in lawful discovery, shareholders through governance or disclosure channels, and journalists when facts become public through litigation, enforcement, leaks, or reporting.

It starts as an internal governance standard. In Decision Accounting, reconstructability means the software can recover the point-in-time governance state in seconds: the rules, authority, decision records, and prior decisions that governed the organization at a particular moment. That is a managerial and software-design claim before it is a legal claim.

It becomes a legal standard when a statute, regulation, contract, listing rule, procurement rule, or court doctrine adopts it and attaches legal consequences to failure. Fuller's point matters because law needs public, prospective, stable rules. Decision Accounting can supply the record architecture, but legal systems decide when that architecture becomes a legal duty.

Calabresi's cheapest-cost-avoider logic is important because it asks the right institutional question: who could have prevented or reduced the harm at the lowest cost? Tort law answers that question after an injury, through plaintiffs, defendants, causation, damages, and remedies. MST and SAPM extend the same logic to system-welfare losses that tort law often reaches poorly: benchmark integrity, ecological capacity, antibiotic effectiveness, fiscal capacity, market trust, or public-health baselines.

SAPM measures the system cost. Decision Accounting records the decision point where that cost was accepted and why the choice was made. Conflictoring can then ask a Calabresi-style question across its seven agents: which actor can most cheaply prevent, reduce, expose, insure, regulate, litigate, or reprice the harm?

The point is to move cheapest-cost-avoider reasoning upstream, before liability arrives and before the system loss becomes harder to reverse.

Decision Accounting can cohere across jurisdictions because the common unit is the decision record rather than a single legal form. Roe's path-dependence means adoption will travel through different vehicles: the UK's Senior Managers Regime, U.S. bank-secrecy exams, EU operational-resilience rules, securities disclosure, procurement rules, or board-governance duties.

The cross-jurisdiction claim is that materially consequential decisions keep needing the same reconstructable elements: who decided, what was decided, when, where, why, under what authority, on what evidence, with what prediction, what communication position, what reconsideration trigger, and what system-welfare consequence, among other fields and record elements. The statute can differ while the record architecture remains comparable.

That comparability lets local legal forms remain different while judgment, authority, events, predictions, outcomes, and welfare consequences can still be studied in a common structure.

Numerical claim withheld pending source admission and independent re-estimation.

It is a diagnostic ratio: annual system-welfare loss divided by annual industry revenue on the same boundary. The dollar numerator combines cited monetary costs, such as cleanup, treatment, enforcement, fraud loss, or fiscal loss, with published shadow prices where the harm lacks a market price, such as mortality risk, carbon damage, or ecological loss.

The point is sign, rank, and control design: identify whether the activity degrades the system, compare severity across domains, and trigger governance, liability, restriction, redesign, or prohibition where appropriate.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

Decision Accounting reduces insider-protection risk through timing, evidence, software checks, tamper-evident records, and independent reviewability. Field 17, system welfare impact, is completed before the outcome is known, so it cannot be backfilled into a flattering story after results arrive. Field 16, prediction, makes the forecast scoreable.

Evidence attachments, authority trails, anti-gaming review, weak-answer controls, and independent access rights, among other controls, make boilerplate and selective memory easier to challenge. The record becomes capture-resistant when those controls push the writer toward a record that is strong under the record-quality standard.

Hansmann's point strengthens the case for Decision Accounting. If owners are the residual claimants, they should want a record system that protects and grows residual value.

Decision Accounting gives owners that tool. Showing your work improves your work: before approval, the firm has to state who, what, when, where, why, authority, evidence, Field 16, prediction, and Field 17, system welfare impact, among other fields. That record can improve the decision while it is still open to revision, expose weak reasoning, force alternatives into view, and identify system-welfare losses that may later return as lawsuits, regulation, insurance costs, reputation damage, financing constraints, or lost operating capacity.

The owner-value case is direct: better decisions, lower reconstruction cost, reduced litigation and regulatory risk, stronger governance evidence, and the possible cost-of-capital benefit the Decision Accounting papers estimate at up to about 200 basis points when the system is implemented well. The enterprise-learning-loop case is also direct: the owner receives structured judgment data showing decisions, non-decisions, events, predictions, outcomes, and welfare consequences in one timeline.

That record can support management, training, risk review, and AI-assisted decision analysis. Ownership supplies the residual claimant. Decision Accounting gives that claimant a disciplined way to protect and grow enterprise value by pushing consequential decisions toward the record-quality standard.

IFRS S1 and S2 operate at a different scale and for a different reporting user. IFRS S1 asks for sustainability-related financial disclosures about risks and opportunities that could reasonably affect the reporting company's cash flows, access to finance, or cost of capital. IFRS S2 applies that investor-useful disclosure model to climate-related risks and opportunities. SAPM measures system welfare at the domain or industry level: what annual system-welfare loss did the activity impose, and how large is that loss compared with annual industry revenue?

That makes S1 and S2 a subset of the system-welfare picture, and usually a smaller one, because investor-material disclosure captures the slice of sustainability or climate risk that matters to the reporting firm's capital-market users. The welfare ledger measures the wider loss to climate stability, ecological capacity, public health, infrastructure resilience, institutional trust, and other system conditions, including losses that may not yet be material to a single reporting company.

Decision Accounting then asks the decision question: who made the decision, what was decided, when, where, why, on what evidence, with what Field 16, prediction, and with what Field 17, system welfare impact, among other fields? The enterprise-learning-loop value is that sustainability disclosure can be connected to structured judgment: capital allocation, product design, supply-chain choices, risk controls, transition plans, event records, predictions, and later outcomes. ISSB disclosure supplies company-level sustainability evidence.

SAPM supplies the domain-level welfare measurement. DA connects both to the decision timeline.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

NIST and ISO 42001 help an organization manage AI risk. NIST AI RMF organizes the work around govern, map, measure, and manage. ISO 42001 supplies an AI management-system standard. Those frameworks still leave a Decision Accounting question: if an AI system scores decision records, what exactly is it scoring, who approved the scoring rule, what weak answers does the software reject, when may a human override the score, and how are scoring errors learned from over time?

A record-scoring AI is itself a governance actor because it changes which explanations pass, which records are flagged, and which decision-makers are pushed to improve their work. Decision Accounting therefore applies the record discipline back onto the AI scorer: who, what, when, where, why, authority, evidence, Field 16, prediction, Field 17, system welfare impact, acceptance criteria, override rules, anti-gaming checks, and outcome feedback, among other fields and controls. The same point applies to AI-assisted decision support more broadly.

AI becomes more useful when the organization has structured judgment data: decisions, non-decisions, event records, predictions, outcomes, and review triggers tied to named authority and evidence. The self-application rule is simple: an AI system that grades governance records must itself be governed by a decision record.

Decision Accounting adds the decision record behind anti-money-laundering controls. FATF supplies the global AML/CFT/CPF standard: anti-money laundering, counter-terrorist financing, and counter-proliferation financing. Its recommendations use a risk-based approach, customer due diligence, monitoring, suspicious-transaction reporting, supervision, enforcement, and international cooperation.

DA targets the upstream choices that determine whether those controls work: who approved the monitoring threshold, why the AML budget was set at that level, which products or customer channels were covered, which were left outside the control system, what staffing and escalation limits were accepted, what Field 16, prediction, was made about the residual risk, and what would require reconsideration, among other fields and controls. Suspicious-activity reports document events after the monitoring system sees them.

Decision Accounting documents the governance choices that determine what the monitoring system is capable of seeing in the first place. The enterprise-learning-loop value is that AML no longer lives as scattered reports after the fact. Budget decisions, product launches, monitoring rules, alert backlogs, regulatory events, suspicious-activity outcomes, missed triggers, and later enforcement findings become one timeline the bank can learn from before the next failure.

This is a proposed disclosure architecture, not a current FASB or IASB requirement, accepted accounting standard, or peer-reviewed result. Financial statements are company-level records. SAPM is usually a domain- or industry-level measure.

Numerical claim withheld pending source admission and independent re-estimation.

A serious future disclosure should also include the game change plan: what the company is doing inside its own operations, what industry-wide game change would reduce the loss, which lawful collaboration or policy channels are needed, and what milestones would show progress. Decision Accounting would add the company-level bridge: which decisions produced the revenue, which decisions accepted or reduced the welfare exposure, what prediction was made, what event later tested the decision, and what outcome followed.

Reform Pathfinder and Policy Lab would supply the game change roadmap. Conflictoring would identify the agents whose independent incentives can help move the industry: boards and CEOs, employees and whistleblowers, regulators, policymakers, shareholders, plaintiff litigators, and communities and the affected public where lawful access exists.

The accounting contribution would be a measurable contra account paired with a governance plan: financial statements show the company's money results, and the welfare disclosure would show the system-welfare cost associated with the industry activity that produced those results.

A Decision Accounting record becomes an audit target when there is a defined assertion and a standard for testing it. The first assertion is existence: did the record exist at the decision time, with timestamping and tamper-evident integrity? The second is completeness: did the software require acceptable answers for who, what, when, where, why, authority, evidence, Field 16, prediction, Field 17, system welfare impact, and the other required fields and controls?

The third is support: do the evidence attachments, authority trails, employee inputs, and Chief Decision Officer review provide sufficient appropriate evidence for the record language? The fourth is performance: what do the Brier scores, weak-answer flags, anti-gaming reviews, override logs, and reconsideration-trigger tests show over time? The fifth is disclosure consistency: does the public or financial-statement disclosure match the internal decision record where disclosure risk is material?

The enterprise-learning-loop audit question is whether the system links the decision to the later event and outcome that tested it, so auditors can sample the full chain from decision to trigger to review to result. PCAOB-style audit standards would still have to define assurance procedures, materiality, sampling, independence, and evidence rules. DA supplies the audit target: the decision record, the scoring system around it, the linked decision-event-outcome timeline, and the governance controls that produced it.

OECD principles say what corporate governance should achieve: accountable boards, reliable disclosure, risk oversight, shareholder rights, stakeholder recognition, and fair treatment of investors. Decision Accounting is a proposed implementation system for that agenda. It gives boards and managers a disciplined way to show how consequential decisions were made, what evidence supported them, what risks were accepted, what prediction was made, how communication was handled, and how system welfare was considered.

It also gives the firm structured governance data: judgment records connected to non-decisions, events, predictions, outcomes, weak-answer patterns, and system-welfare consequences. The proposed primary market test is cost of capital: measure cost of debt, cost of equity, and weighted average cost of capital before DA adoption, then track whether the measures decline after implementation across many adopters.

Audit and governance indicators should be treated as mechanism evidence: record completeness, prediction accuracy, weak-answer rates, anti-gaming flags, disclosure consistency, governance-state recovery time, board-review quality, restatements, litigation events, regulatory investigations, regulatory fines, enforcement actions, and other negative outcomes. DA is not an OECD standard today; it is a candidate record architecture that regulators, firms, auditors, investors, and standard-setters could evaluate against existing governance principles.

A Chief Risk Officer should evaluate Decision Accounting by governance economics: does the record improve risk decisions enough to justify the record burden? The answer is strongest where risk teams already maintain scattered inventories, approvals, model documents, controls, committee minutes, and audit evidence.

DA turns those fragments into dated decision records tied to Field 16, prediction, Field 17, system welfare impact, evidence, authority, weak-answer controls, outcome scoring, reconsideration triggers, and tamper-evident integrity, among other fields and controls. The immediate value is better decisions before they harden. The record forces a choice to meet the record-quality standard while the choice is still open to revision.

The second value is reconstruction: after a bad outcome, audit, regulatory exam, board review, or lawsuit, the organization can recover the point-in-time governance state in minutes: the governing rules, the decision record, the evidence and authority used, the prediction made, and the trigger that should have caused reconsideration.

The third value is decision intelligence. In an AI-era firm, documents, messages, and dashboards are useful, but the highest-value internal data is the structured record of judgment: what the organization decided, why, who owned the choice, what was predicted, what event later tested the prediction, and what outcome followed. Decisions, non-decisions, event records, warnings, prediction scores, outcomes, and weak-answer patterns become a decision-support dataset.

That is the dataset a risk organization needs to see cause and effect across time: which warnings mattered, which choices were missed, which predictions failed, and which decisions should be changed before the next loss.

Numerical claim withheld pending source admission and independent re-estimation.

The governance answer is privilege bifurcation, process discipline, and a game change plan. The record should separate legal advice from operational reasoning, control disclosure pathways, and show who, what, when, where, why, authority, evidence, uncertainty, Field 16, prediction, communication, review triggers, residual-risk acceptance, and the planned game change response, among other fields and record elements.

The game change plan should cover the firm's own decisions and the industry-level work needed to reduce the loss, using lawful channels such as Reform Pathfinder, Policy Lab, Conflictoring, regulator engagement, standard-setting, and antitrust-counsel-supervised collaboration where collective action is required. A record that meets the record-quality standard and is tied to a credible game change plan can defend a serious process.

A welfare admission without a game change plan exposes an unmanaged process.

The implementability question has a public answer: the seventeen-field record has been implemented, filled, and machine-scored at scale, and the evidence is in the program’s own replication study, which is public.

Start with the standard itself. The Decision Record & Control Standard is published under an Apache 2.0 license as part of the Decision Accounting paper: field definitions, the data model, the cryptographic integrity architecture, and gaming-detection parameters, specified precisely enough that an engineer who was never in the room can build a conforming implementation from the document alone. That is the difference between a specification and a principles framework, and it is the engineering version of the five-minute test.

Then the demonstration. In the program’s replication of Calvano, Calzolari, Denicolò and Pastorello (2020)⁶⁹ — the experiment where pricing algorithms teach themselves to collude without communicating — every algorithm files the complete seventeen-field record at every pricing period: who, what, why, evidence, the Field 16 prediction, the Field 15 alternatives it considered and rejected, and the Field 17 system-welfare consequence, among the rest of the seventeen. Millions of records per run, each machine-scored for completeness against the published standard. What the replication tests is the record format and the audit mechanism, not the human difficulty of answering the questions. In the laboratory the hard fields are easy by construction: the environment is fully specified, so the system-welfare consequence is computable exactly, the alternatives are the finite set of prices the algorithm weighed, and the prediction is read directly from its own value estimates. A human filling Field 17 for a real decision faces a genuinely harder problem — naming the affected system, estimating the damage without double counting, matching the numerator to a revenue denominator, and defending the measurement boundary. The Decision Accounting paper addresses that burden directly: where a calibrated estimate does not exist, the field requires a structured qualitative entry — the named system, the direction and range of the effect, and the horizon — not a finished welfare study. The replication shows the discipline can be enforced. It does not show the answers come cheap, and the framework does not claim they do. The replication package — environment, seeds, code, and outputs — publishes with the paper, so the demonstration can be rerun by anyone.

The completed conditions support two results; the audit conditions are still in progress. Record-keeping that no one audits leaves collusion unchanged from baseline, while the algorithms’ own Field 16 predictions become nearly perfectly calibrated. Honest paperwork, unchanged conduct — exactly what the theory predicts when records have no readers.

What no study has yet proved is organizational: sustained adoption inside institutions, examiner acceptance across a live supervisory cycle, and decision-quality evidence in adopters. Those are stated in the Decision Accounting paper as falsifiable propositions with study designs, not assumed.

A framework that must be interpreted can die in translation. A specification a machine can verify cannot.

The program is substantially market-liberal in design because most of its game-change machinery runs through private or market-facing actors. Conflictoring has six recurring agents: employees and whistleblowers, CEOs and boards, regulators, policymakers, shareholders, plaintiff litigators with lawful access in the relevant setting, and communities and the affected public. Only one of those six is the regulator.

Numerical claim withheld pending source admission and independent re-estimation.

The point is to make the missing system-welfare cost visible so market actors can respond: investors can price risk, insurers can price coverage, boards can change strategy, employees can escalate information, litigants can test liability, and customers can change demand. Reform Pathfinder and Policy Lab then ask which lawful game change can reduce the loss. Regulation remains one lane, but the architecture is built around measured information, private incentives, and institutional competition rather than a central planning command.

The starting point is corporate-record governance. A decision record is an internal company record, and often valuable organizational intellectual property, even when it names employees, managers, engineers, or the Chief Decision Officer. Once timestamped, the original record is permanent.

It cannot be edited, corrected, modified, redacted, or anonymized; a later clarification, correction, or reversal must be made as a new timestamped record linked to the original. Privacy principles associated with consumer-facing data still matter, but they do not give an employee a general right to delete a truthful corporate governance record. The company owns its records and sets the retention schedule through legal and governance policy.

That said, employee personal information inside the record is still regulated. In EU settings, the right to erasure is real but qualified: organizations may retain personal data when retention remains necessary for lawful purposes, legal obligations, or the establishment, exercise, or defense of legal claims.⁷⁰ Protection therefore comes from scope, access control, legal supervision, and adopter-specific operating rules: materially consequential decisions only, role-limited access, privacy review, retention rules, export controls, and misuse controls.

Numerical claim withheld pending source admission and independent re-estimation.

There are strong reasons to keep decision records for a long period: litigation defense, audit, regulatory review, governance learning, intellectual property, AI-assisted training, and comparison of decisions, event records, predictions, outcomes, and weak-answer patterns over time.

The Chief Decision Officer, preferably with legal training or close legal supervision, should work with legal, privacy, human resources, compliance, and the board to fit retention, access, export, worker-protection, appeal, and misuse rules to the company's culture, precedent, employee handbook, existing policies, employment-law setting, and governance framework.

Numerical claim withheld pending source admission and independent re-estimation.

Numerical claim withheld pending source admission and independent re-estimation.

If adopted, the plan should identify the Conflictoring agents the board and CEO intend to work with, the Reform Pathfinder route, the Policy Lab route, milestones for reducing the exposure, and antitrust-counsel-supervised collaboration where industry cooperation is required. A short-seller then has less room to frame the number as a concealed liability because the company has disclosed both the exposure and the plan for reducing it. Short-sellers can use true negative information; that is part of market price discovery.

Numerical claim withheld pending source admission and independent re-estimation.

The current answer has three layers: retrospective case evidence, completed laboratory evidence, and the live-organization field test that still remains.

First, the case evidence. Major failures often require months of forensic reconstruction to answer questions a contemporaneous decision record would have preserved in minutes. Retrospective case work shows how weak rationale, missing authority, thin evidence, poor prediction, and missed review points become visible when the decision is reconstructed. That supports the mechanism and identifies the class of failure DA is designed to prevent.

Second, the Calvano follow-up now supplies laboratory evidence in algorithmic pricing. The implementation ran pricing agents that filed the full seventeen-field decision record at every pricing period. The result is precise: unaudited records did not reduce collusion. The collusion index, Δ, is normalized from 0 to 1: Δ = 0 means competitive-benchmark pricing, and Δ = 1 means monopoly-benchmark pricing. The no-record, no-audit baseline was about 0.82, and unaudited decision records stayed about 0.82-0.83, while Brier forecast error improved from roughly 0.24 early in training to roughly 0.054 late in training. Accurate records alone did not change behavior.

Third, consequences mattered. Adding system welfare directly to the reward cut Δ to 0.2863 ± 0.0194, much closer to the competitive end of the scale than the 0.82 baseline. Audits with a consequence also worked: at a 50% audit rate, forcing the audited agent to charge the competitive benchmark price, p^N = 1.4729, for ten pricing decisions cut after-training Δ to about 0.188. The alternatives-only audit arm clarified the boundary: proof from the rejected-alternatives field created little forced-price exposure by itself and left Δ around 0.82. The lab mechanism is therefore record plus authorized reader plus consequence, with system-welfare scoring strongest when it enters incentives directly.

The cleanest live-organization field test is still a market test. Measure a firm's cost of debt, cost of equity, and weighted average cost of capital before DA adoption, then track whether and how quickly those measures decline after implementation. If many adopters show a large and repeated post-adoption decline, for example near the hypothesized 200-basis-point range, that would be strong evidence that DA is creating market value.

The study should also report announcement-window market reaction because investors may price adoption before operating results appear. Staggered adoption across firms can strengthen the evidence if the econometrics control for market, industry, rate, credit, and firm-specific shocks.

Secondary measures such as record completeness, weak-answer rates, prediction accuracy, governance-state recovery time, audit results, board-review quality, litigation outcomes, regulatory outcomes, and decision-quality scores are mechanism evidence. They help explain why the cost of capital moved. The remaining missing item is completed field evidence from live organizations across a full adoption cycle.

Numerical claim withheld pending source admission and independent re-estimation.

The Calvano follow-up makes the labor-and-climate warning sharper. In the algorithmic-pricing test, complete decision records by themselves did not change conduct. Pricing agents filed the full seventeen-field record every period, and their predictions became accurate, but collusion remained near the no-record baseline when no authorized reader used the record. The collusion index, Δ, is normalized from 0 to 1: Δ = 0 means competitive-benchmark pricing, and Δ = 1 means monopoly-benchmark pricing. In the clearest cells, unaudited decision records stayed about 0.82-0.83 against the no-record, no-audit baseline of about 0.82, while forecast error fell from roughly 0.24 early in training to roughly 0.054 late in training. Accurate paperwork and harmful conduct coexisted.

The result changes the policy lesson. Pricing harm is useful only when it feeds an action channel. The same follow-up shows that audits with a consequence can interrupt collusive pricing: at a 50% audit rate, forcing the audited agent to charge the competitive benchmark price, p^N = 1.4729, for ten pricing decisions cut after-training Δ to about 0.188. The separate alternatives-only audit did not move conduct because it almost never created enough forced-price exposure; it produced proof of what the agent rejected, but proof without a consequence did little inside the lab.

Numerical claim withheld pending source admission and independent re-estimation.