Discovery Is Not Governance

Post-hoc discovery starts after system welfare has already failed

The paper argues that discovery-based accountability is structurally late: it reconstructs decisions after C has degraded instead of requiring decision rationale when the risk-shifting decision is made.

• Missing System Theorem: the firm-regulator game G is defined over A and B, while system welfare C is outside the payoff space.
• Hollow Win: C=0, A=1, B=1, so firm and regulator both satisfy internal metrics while the system degrades.
• Discovery changes the post-failure information set; it does not change the payoff structure that produced the decision.

The Hollow Win is locally Pareto-optimal because C is missing

Inside G= A,B , both parties can view the outcome as successful. The failure appears only after the game is transformed to include C.

• A=1: the firm gains by externalizing risk or reducing compliance costs.
• B=1: the regulator gains by avoiding political friction, meeting narrow metrics, or preserving future options.
• C=0: public health, safety, environmental quality, or systemic stability degrades outside the recorded bargain.
• The paper names this structure Conflictoring: formally adverse parties share incentives to keep damaging rationale outside the observable record.

VW passed certification while real-world NOx ran 10 to 40 times over the limit

The diesel case is the paper's main worked example of a Hollow Win in a regulatory certification game.

• Period: 2008-2015, detected in September 2015.
• A=1: VW avoided about $6.5B in engineering costs and sold 11M vehicles with defeat devices across 8M customers.
• B=1: CARB and EPA certification metrics showed 100% compliance.
• C=0: real-world NOx emissions were 10-40x legal limits; the paper reports 59 premature deaths annually in the U.S. and 1,200 premature deaths in Europe over seven years.
• β_W: $193B in system welfare cost against €202.5B 2015 revenue, about 0.86 to 0.95 depending on exchange-rate treatment.

Discovery loses when externalizing risk pays more than expected detection

1 states that discovery cannot prevent degradation when the expected private gain from risk-shifting exceeds the expected discovery cost.

• Expected gain: E[Gain(D)] = ΔΠ_A + ΔΠ_B.
• Expected discovery cost: E[Cost(D)] = P(detection) x P(attribution) x Penalty.
• A and B reduce detection and attribution by under-disclosing rationale, routing records through privilege, or never creating reconstruction-complete records.
• Even complete discovery of A and B records cannot derive C because A7 says no function f(A,B)=C exists for all decisions.

The legal record splits into sanitized operations and shielded risk analysis

Axiom A3 explains why discovery often finds paperwork without finding the decision calculus that made the harm predictable.

• Operational stream: discoverable, broadly circulated, and stripped of controversial rationale.
• Privileged stream: legal-risk analysis and controversial reasoning shielded from discovery.
• Reconstruction Gap: the post-hoc record lacks who knew what, which alternatives were rejected, and why the risky option looked acceptable at the time.
• Corollary 1.2: discovery remains incomplete when the actual decision calculus is in the privileged stream.

Field 17 puts system welfare into the decision record

2 claims that mandatory Decision Accounting changes G into G' by requiring a contemporaneous ledger for decisions meeting the Material Systemic Risk threshold.

• Field 1 identifies the decision-maker.
• Field 5 captures the reasoning.
• Field 6 records the evidence base.
• Field 12 records alternatives considered.
• Field 17 records projected system-welfare consequences, which cannot be inferred from A and B payoffs alone.

A DA record must pass the Stranger Test

A reconstruction-complete record must let an independent auditor understand the decision without access to the original decision-maker.

• Completeness: the record contains the information needed to judge the system-welfare criterion.
• Interpretability: an independent auditor can read the record without private context.
• Timing: the record is created contemporaneously with the decision, not after litigation begins.
• Without those three conditions, the paper classifies accountability as archaeology.

Decision Accounting makes detection and attribution equal one

The paper's mechanism is not more disclosure volume. It is a mandatory record that makes C visible, attributable, and non-routable.

• In G', E[Cost'(D)] = P(detection') x P(attribution') x Penalty' + Reputational_Cost + Regulatory_Cost.
• P(detection')=1 because the DA record exists and is auditable.
• P(attribution')=1 because the decision-maker is identified in the ledger.
• Field 17 is treated as a public-interest record, so risk-shifting rationale cannot disappear into privilege.

Patterns of harm do not identify the decisions that caused them

3 rejects reform based only on aggregate evidence of degradation because aggregate harm does not recover decision rationale.

• E_cross can show that C degraded by ΔC over period T.
• It cannot show which D_i caused the degradation, what reasoning supported D_i, what alternatives were rejected, or whether decision-makers knew C would degrade.
• Four failure modes follow: overinclusion, underinclusion, delay, and capture.
• The Archaeology Trap repeats the same cycle: observe degradation, investigate, attribute imperfectly, reform weakly, observe new degradation.

BP and Purdue fit the same missing-ledger pattern

The paper applies the model beyond Volkswagen to Deepwater Horizon and opioid marketing, with the common feature that decisive risk-shifting rationale was not captured contemporaneously.

• BP Macondo: the paper treats Deepwater Horizon as a case where discovery came after collapse rather than governing the safety decisions beforehand.
• Purdue Pharma: using annual revenue of about $3B, the paper reports β_W≈500, which it classifies as strong intractability.
• Across the three cases, the key defect is absence of R(D): no recoverable ledger of reasoning, alternatives, evidence, and projected system-welfare consequences.

Norwegian aviation reporting records safety decisions before litigation reframes them

The paper identifies the Norwegian Civil Aviation Authority's mandatory occurrence reporting system as the closest analogue to Decision Accounting.

• The system requires contemporaneous recording of safety-relevant decisions.
• Its mechanism is separation of record creation from the enforcement game.
• The falsification section treats strong no-blame aviation reporting cultures, including Norway and New Zealand, as candidate tests for whether discovery alone can ever sustain (1,1,1).

The paper names five observations that would weaken the theory

The model is framed as falsifiable rather than as a policy preference.

• Discovery sufficiency: a domain sustains (1,1,1) through discovery alone across multiple decision cycles.
• Privilege irrelevance: risk rationale stays in operational records, privilege routing does not occur, and C remains preserved.
• Cross-sectional reform success: reform based only on E_cross produces sustained system-welfare improvement.
• Non-derivability violation: a function or model predicts C from only A and B payoff data with accuracy above 0.95.
• Conflictoring counterexample: A and B have opposed disclosure incentives and produce sustained (1,1,1) without DA.