Risk aversion changes the decision criterion

The expected monetary rule used in the worked studies is a declared objective. It is not a universal account of institutional preference. A committee may care about the distribution of outcomes, liquidity in an adverse state, a mandate constraint or an exposure that cannot be diversified. Those concerns can make an action with a lower expected monetary payoff preferable under the institution's actual decision rule. The analyst should identify the disagreement as an objective or admissibility question instead of insisting that correct expected-value arithmetic settles it.

Risk aversion concerns how consequences are valued, not merely how uncertain they are. A probability expresses a belief about a represented state. A utility function would express a preference over its consequences. Changing the prior to make a risky action less attractive is an inappropriate way to insert risk aversion when the belief itself has not changed. The institution should preserve the probability evidence and state the preference or constraint separately. This keeps the model's epistemic assumptions from becoming an instrument for achieving a desired recommendation.

The binary app has no utility-function input. It therefore cannot calculate expected utility, a certainty equivalent or an optimal action under a chosen nonlinear preference. The paper discusses these methods conceptually and uses the monetary outputs as a reference comparison only. A real utility analysis would need a justified consequence measure, a preference elicitation process and appropriate calculation. An analyst should not apply an arbitrary haircut to the monetary result and call that a calibrated risk-aversion adjustment. Such a change is another supplied payoff assumption unless its method is actually defined and supported.

Constraints provide a different way to represent some institutional concerns. An organisation may rule out an action that exposes it to an unacceptable service interruption or breaches a mandate, regardless of average payoff. This is an admissibility condition rather than a smooth preference penalty. The action should be excluded from the operative menu or reviewed with a model that explicitly handles the constraint. The restricted-action fixture illustrates the effect of removing a response but does not establish why any real response should be prohibited. That decision belongs to the relevant authority and evidence.

The distinction matters for information value. An instrument that enables a high monetary payoff through an inadmissible action can have an overstated apparent benefit. An instrument that reduces exposure under a utility objective can have a different value from its risk-neutral monetary gain. The current paper cannot infer either value from the existing engine. It can identify which assumption would change and what additional method is required. The committee should avoid translating an attractive monetary EVSI directly into a research budget when its actual objective has not been reconciled with the model.

Expected regret offers another transparent reading of the same supplied matrix. The engine calculates state-specific regrets and their probability-weighted expectation for each action. The current monetary optimum minimises expected regret in this fixed menu because the perfectly informed benchmark is the same across actions. This equivalence does not establish a maximum-regret rule or a psychological measure of regret. A maximum-regret decision would focus on a different aggregation of state consequences and need an explicitly declared method. The state-regret table allows readers to inspect the inputs without pretending that the app has selected that alternative rule.

The illustrative regret table reveals different exposure patterns. Commit gives up no payoff relative to the best action in success, but has a large failure-state regret. Hold avoids failure-state regret while giving up the success upside. Stage divides the sacrifice differently. These are state-contingent consequences of the supplied matrix. A committee can use them to discuss its mandate and tolerable exposure, but it should not infer a universal preference for the middle action or the largest success payoff. The selected rule and evidence perimeter determine what the table means for the institution.

Risk capacity also differs from risk preference. An institution might be willing to accept a downside but unable to fund the cash consequence when it occurs. A present-value payoff can conceal that timing problem. Another institution may have capacity but a mandate that restricts the exposure. The binary payoff engine does not calculate dated liquidity, funding covenants or legal authority. A decision package should therefore retain a separate capacity and permission review when these conditions are relevant. The monetary result should not be used to imply that a funding path exists.

The committee should ask whether exposures are repeated and diversifiable. A linear monetary expectation can be a useful planning reference for a set of limited choices, but concentration and common causes can change the institution's capacity. The app does not model a portfolio or dependence across opportunities. If the research commission supports a single consequential commitment, an aggregate expectation should be accompanied by a review of the adverse state and response. The paper offers no empirical conclusion about how a particular institution should trade average return against concentration.

Preference elicitation requires its own governance. The person supplying a utility or constraint should have authority to represent the institution, and affected interests should be considered through the appropriate process. An analyst's convenient parameter is not automatically an institutional preference. The decision record should distinguish a provisional sensitivity assumption from an approved objective. If participants disagree, the report can show the consequences of alternative declared rules while preserving the unresolved judgement. Hiding the disagreement inside an adjusted payoff cell makes the result harder to challenge.

Research can itself help clarify preferences, but that is a different information problem. A workshop may reveal how decision owners value consequences or what constraints actually bind. This can improve the decision architecture without observing the opportunity's success state. The binary instrument fixtures do not calculate the value of eliciting institutional preferences. A team should describe that purpose directly and avoid borrowing the commercial state's EVSI as a budget justification for every governance or deliberation activity. The information object, decision and response mechanism should remain explicit.

A claim of prudent conservatism also needs a basis. Lowering all upside inputs, raising every downside and imposing a decision margin may duplicate the same concern several times. The result can become systematically cautious without a traceable risk rule. An evidence package should record which uncertainty is represented through probabilities, which consequence uncertainty is represented through payoff sensitivity, and which institutional preference is represented through an objective or constraint. This prevents a plausible cautionary label from replacing a coherent decision method.

Use an objective-change audit when the selected action differs from the monetary leader. A hypothetical committee can exclude an action because it exceeds a protected exposure, choose a supported utility rule because concentrated loss matters, or apply another authorised criterion. Each route changes a different part of the decision architecture. Removing the action changes the menu; utility changes valuation; another criterion adds a separate comparison. None should be hidden by rewriting a payoff cell until the monetary output matches the committee's preference. Preserve the monetary reference and identify the actual route. A remaining difference then explains the cost or consequence of the chosen rule instead of appearing to be an arithmetic inconsistency.

Preference disagreement also needs a different evidence request from payoff disagreement. If hypothetical reviewers agree on the consequences but disagree about accepting the downside, more precise cash estimates may not resolve their risk preference. If they disagree about whether the institution can fund that downside, the question is capacity and admissibility. If they disagree about the downside itself, the consequence evidence is relevant. The objective register should route each dispute accordingly. This is a practical use of the boundary around risk-neutral money: it prevents the team from purchasing a predictive signal to resolve a normative or financing question that the signal cannot answer.