Breaking conformism through multiple audiences

The Decision Accounting as a Report-Incentive Mechanism paper models the reporting game formally. A decision-maker bears a cost kappa to truthfully record the welfare impact W. A misreport is detected after the fact with probability p by a Conflictoring auditor, the seven-lane protocol that imposes review costs simultaneously. On detection, the decision-maker bears a liability L. Truthful recording is a best response when and only when pL >= kappa, the paper's Proposition D.1: the expected sanction for lying must at least cover the cost of telling the truth. The unpredictable-reader property enters through the detection probability p. When the audience is known and singular, the decision-maker can estimate p accurately and tune the report to sit just under the detection threshold. When the audience is plural and unpredictable, p stops being a single number the writer can estimate and becomes a distribution over possible readers. The decision-maker can no longer compute the expected sanction for a misreport, because they cannot predict which audience will read the record or what that audience will be able to detect. That uncertainty is what separates this mechanism from a Myerson-Satterthwaite private-value impossibility result (no mechanism guarantees efficient trade⁷, truth-telling, and no subsidy at once): the obstacle is not unobservable private valuations but a verifiable report whose reader cannot be pre-selected.

The information-exclusion paper argues that⁸ the Missing System Theory (MST) is, at root, an information-exclusion result. The system-welfare coordinate W is absent from the disclosed records that firms and markets report, by the same construction that keeps W out of the payoff space firms optimize over. That off-ledger status creates a specific informational friction. Because the welfare measure is not part of any mandatory or customary disclosure, it is not reflected in equilibrium prices. The same friction explains two findings at once: there is no ex-ante premium for the coordinate (no price paid for it in advance), and there is a positive rent to the first party that does disclose it. Decision Accounting is the disclosure event that carries the coordinate onto the ledger for the first time. The measurement claim is deliberately modest in form: it is rank and sign based rather than level dependent, meaning the framework commits to the direction and ordering of welfare effects rather than to precise dollar levels. This is the information-theoretic reason the multi-audience mechanism has anything to work on. The coordinate was never in the ledger, so any disclosure that places it there changes the information environment structurally rather than marginally.

The Evidence That Decision Accounting Works paper⁹ (chapter 9's six-case evidence base) is what the falsification condition below rests on.

> *A composite scene, constructed from the protected Chapter 18 mechanism; it names no organization > and makes no empirical frequency claim.* > A manager drafts the memo that will justify shipping a product on schedule despite an unresolved > safety signal. She knows exactly who reads it: the risk committee, the same seven people who have > approved every proposal put in front of them for three years. So she writes the memo they reward > — confident tone, the signal noted and "being monitored," the schedule preserved. It sails > through. The record is clean, complete, and signed. > > Two years later the same memo is read by three people she never imagined: a regulator building an > enforcement file, a plaintiff's lawyer in discovery, and the successor board trying to work out > what the last one knew. To them the memo reads as a warning that was buried. > > She did not lie to a stranger. She told the truth to the only reader she could see — and that is > the problem this chapter is about. *(Narrator throughline, carried in from every prior handoff: the missing ledger line, and the recurring question — *what was decided, who gained, what system carried the residual, what changes the next decision? this chapter asks a fourth thing: who gets to read the answer.)

The accountability paradox Protected Chapter 18 uses Andrea Prat's 2005 model to show how transparency to a known evaluator can induce conformism when the evaluator rewards a preferred action and the decision-maker cares about the evaluator's inference or career judgment. The decision-maker then has an incentive to choose or present what the evaluator expects. This is an equilibrium result under the model's information and career-concern assumptions. It does not claim that every known reader always causes dishonesty. Its curriculum use is a design warning: adding disclosure to one predictable evaluator can strengthen the incentive to satisfy that evaluator without improving the record's truthfulness. Source boundary. This is the protected Chapter 18 model claim, located at chapter-18 blocks[0].sections[0–4]; the paper-level assumptions, version, and external review receipt remain pending. the chapter teaches the stated mechanism and its limits under those assumptions. Why the patches fail in the bounded comparison MODEL COMPARISON (protected curriculum) The protected Chapter 18 comparison treats Prat's result as a constraint to route around. In the bounded comparison, each proposed fix keeps the same structure — one decision-maker reporting to one evaluator — and tries to patch it: | Patch | What it did | Why it failed | | Information restriction | Hide the record from all but a trusted auditor | The auditor becomes the single known audience; Prat's logic returns intact | | Audience rotation | Change evaluators period by period | Evaluator-specific conformism re-forms within each period | | Anonymous evaluation | Hide the evaluator's identity | The writer infers the evaluator's type from context; anonymity leaks | | Incentive contracts | Pay for truth-telling | The honest counterfactual is never observed, so the contract can't condition on it | | Reputation systems | Let a track record build trust | Reputation is the conformism mechanism — you earn it by giving readers what they want | One structural reason unites the failures: a single-audience game cannot be repaired from inside the single-audience game. The structure has to change. The multi-audience mechanism — the three conditions The rule change is an audience-expansion step: expand the readership from one known evaluator to several principals whose criteria are uncertain and divergent. The transformed game has a different equilibrium — conformism becomes hard to sustain, because you cannot optimize for a reader you cannot predict. The argument rests on three conditions, and all three must hold at once: 1. Audience plurality — at least two principals make *independent first-pass judgments on the same frozen record simultaneously*. Sequentially disclosing a revised record to one reader at a time is disallowed because it lets the writer retune the record for each reader. A sequential follow-up investigation is compatible with the design after those first-pass judgments are frozen: it develops evidence or resolves disagreements without replacing the independent judgments. 2. Objective divergence — the principals apply structurally different criteria. A regulator wants risk disclosure; a board wants strategic justification; the plaintiff bar wants welfare documentation. Satisfying one tends to mean failing another — unless the record is simply accurate, full, and candid about what was uncertain. 3. Audience uncertainty — the writer cannot predict which principal's evaluation will turn out to be decisive, and the record must survive readers who do not yet exist: future litigants, successor boards, later regulators. When all three hold, the writer has to produce one record that survives regulatory, board, litigation, capital-market, and later historical review at the same time. The design therefore aims to preserve the point-in-time governance state and the actual reasoning in one record; whether it does so in practice is an empirical test. Does this solve Prat? The claim is a hypothesis, not completed field evidence: the multi-audience architecture may dissolve Prat's dilemma, because the writer can no longer optimize for one known, friendly reader. When the same record may later be read by auditors, regulators, boards, compliance officers, legal reviewers, successor managers, and authorized outside reviewers — parties with different powers and incentives — the best available strategy becomes to make the record accurate, defensible, evidence-grounded, and honest about tradeoffs. Whether that actually changes behavior in real organizations is the open field test, not a settled finding. The stranger test — the mechanism as a procedure you can run The mechanism has an operational form simple enough to run in a room: put the decision record in front of someone who was not in it. Give them a short reading and ask whether they can recover the point-in-time governance state. The six required outputs are: (1) what was decided, (2) who decided it, (3) the actual reasoning, (4) the authority used, (5) the governing rules, and (6) what would trigger reconsideration. Any missing output is a failed stranger test pending an explanation and a linked correction. The test's power comes from the same property as the formal mechanism: an unfamiliar reviewer still has preferences and evaluation criteria, but the writer cannot know in advance which criteria, questions, or conflicts that reviewer will bring. The record therefore cannot be reliably pre-shaped to please one predictable reader. A pass by an unfamiliar authorized reviewer is evidence that the reasoning can be recovered under that test; it does not establish a universal result. *(What that record contains — the seventeen fields, including SYSTEM WELFARE, one of the seventeen — is the Decision Accounting chapter's home. this chapter only asks who reads it.)* The privacy conditions — plurality without surveillance Plural readership is powerful, which is exactly why it can go wrong. Two objections have to be answered inside the design, not waved off: - Structural-hole power (Burt). Does the record hand power to whoever controls what gets written down? A properly built record reduces that brokerage power rather than creating it: it is attributable, contemporaneous, tamper-evident, and reviewable by defined audiences with different powers — so the reasoning no longer lives inside one privileged channel that a broker controls. - Surveillance drift (Zuboff). Does the record system slide into behavioral-control infrastructure? The safeguard is governed scope and governed access: record materially consequential organizational decisions, not ordinary employee behavior. Access, retention, export, derived datasets, worker protection, and personal-exposure rules are designed by the adopter under legal supervision. A plural audience with ungoverned access is not accountability; it is monitoring wearing accountability's clothes. These are design conditions. A governed-access plan must state: 1. Scope: record materially consequential organizational decisions and exclude routine worker observation. 2. Roles and purposes: each reader receives access for a named legal or governance purpose; access is logged and reviewed. 3. Correction: preserve the original timestamped governance content and attach a linked correction or contestation record. A person may challenge inaccurate personal data without erasing the decision history. 4. Lawful redaction: redact or mask a reader's view when law, privilege, safety, or personal-data protection requires it. Record who authorized the redaction and preserve a tamper-evident link to the protected source record. 5. Retention and deletion: assign a lawful retention schedule by record class. Where law requires deletion or restriction of personal data, retain only the minimum governance proof permitted and record the legal action; do not use “append-only” as a reason to ignore governing law. 6. Derived data: allow anonymized or aggregated datasets only under purpose, export, re-identification, and worker-protection controls. Plural readers do not need identical visibility into personal data. They need access to the same governance truth within their lawful permissions. If the organization can show substantively different reasoning to different audiences, sequential optimization returns and the mechanism weakens.

the chapter's output is a reader/access plan for a real decision: three audiences that satisfy plurality + divergence + uncertainty, plus an access rule that keeps them independent without becoming surveillance. - Worked — a pharmaceutical safety decision (ship-or-hold on an unresolved adverse signal). Audiences: (a) the drug-safety regulator — rewards complete adverse-event disclosure; (b) the board risk committee — rewards strategic and financial justification; (c) patient/plaintiff counsel under lawful discovery — rewards welfare documentation. The criteria genuinely diverge, so no single slant satisfies all three; the writer can't know which reader becomes decisive (approval today, litigation years out). Access rule: regulator and board read in full and contemporaneously; plaintiff counsel only under a litigation hold; individual employees named in the record get worker-protection handling; no export into behavioral-scoring datasets. - Faded — an aviation maintenance deferral. The three audience slots are given (safety regulator; the airline's operations board; injured-party counsel). You fill in what each rewards, why the decisive reader is unpredictable, and one access rule that keeps plurality real. - Independent (transfer) — pick your own high-risk decision. Name three audiences with divergent criteria, state why the writer cannot predict which one reads decisively, and write one governed-access rule that preserves plurality without turning the record into staff surveillance. <details><summary>Complete answer key for the faded and independent work</summary> Faded aviation answer. The safety regulator evaluates compliance and operational risk; the operations board evaluates safety, continuity, and resource consequences; injured-party counsel, if lawful access later arises, evaluates causation, knowledge, and harm. The decisive reader is uncertain because the decision may remain internal, trigger supervision, or become part of later litigation. A passing access rule gives the regulator and board contemporaneous role-based access, places later legal access under discovery and protective-order rules, masks unrelated employee data, logs every access, and preserves linked corrections. Independent answer standard. Full credit requires at least three genuinely different criteria, simultaneous access to one governance record within lawful permissions, an explanation of reader uncertainty, and rules covering scope, role, purpose, correction, redaction, retention, and misuse. A plan fails if all readers report to one principal, if the writer can prepare different substantive versions, or if routine employee behavior becomes the subject of the record. </details>

> A firm decides to defeat conformism by disclosing every material decision, in full, to its primary > regulator. Does routing more disclosure to that one powerful known audience solve the Prat > problem? > *(a) Yes — full transparency to the regulator removes any incentive to shade the record · > (b) No — a single known audience, however powerful, is the exact condition Prat's result > describes.* > > (b). If you picked (a), you're carrying the misconception the whole chapter dismantles: that > transparency and accountability are the same thing. They are not. Disclosure aimed at one > known reader is precisely what Prat shows produces conformism — the writer learns what that reader > rewards and reports it, more efficiently now that the channel is formal. The mechanism is not the > volume of disclosure but its unpredictable plurality under governed access: several readers, > divergent criteria, no way to know which one is decisive, and no selective back channel.

> So the fix is a plural, unpredictable readership with real power to examine the record. But real > power is exactly what incumbents capture — they buy the regulator, starve the reviewer, and the > plurality collapses back to one friendly reader. the Conflictoring chapter: where durable, capture-resistant > audiences actually come from — the seven lanes, and how a captured one gets reallocated¹².

| Depth card | Tier | Minutes | Prerequisite | Learning objective | Claim status | Evidence card | Return link | | Reporting incentives and the unpredictable-reader property | formal depth | 30 | the Decision Accounting and multi-audience chapters core | State the conditions under which expected review consequences can support truthful recording | proposed model | M08-E1 | the multi-audience chapter multi-audience mechanism | | Five single-audience patches | evidence depth | 25 | the multi-audience chapter core | Diagnose why a patch retains a predictable evaluator | model comparison | M08-E2 | the multi-audience chapter misconception check | | Governed access in full | implementation depth | 35 | the Decision Accounting and multi-audience chapters core | Draft scope, role, purpose, correction, redaction, retention, export, and worker-protection rules | design proposal | M08-E3 | the multi-audience chapter independent task | | Structural holes and record brokerage | formal/evidence depth | 20 | the multi-audience chapter core | Test whether control of the record recreates a privileged broker | design proposal | M08-E4 | the multi-audience chapter stranger test | | Information exclusion | formal depth | 20 | the theorem chapter | Connect independent readership to the separately recorded system coordinate | proposed theorem; owned by the theorem chapter | M03-E4 | the multi-audience chapter forward bridge |