Payoffs must share a decision basis

A payoff matrix is an account of consequences within an agreed perimeter. It is not merely a convenient set of numbers for an expected-value calculation. Every cell should specify whose consequences are represented, when they occur, what baseline they use and which material effects are excluded. The engine can compare supplied monetary cells reliably without knowing any of these facts. The committee must therefore review the perimeter before trusting the ranking. A correct multiplication cannot correct a payoff that omits the person bearing the downside.

Incremental payoffs require a common baseline. If Commit includes the cost of the investigation but Stage omits it, the comparison confounds the action with an inconsistent accounting boundary. If a historical sunk cost is charged to one action but ignored for another, the table can reward a choice for an expense that no present decision can avoid. The useful-signal fixture supplies state payoffs before the proposed test cost and treats that cost separately. This is an explicit modelling convention, not a claim about an actual accounting system or a universal rule for every research programme.

The currency unit also needs interpretation. The paper labels the illustrative payoffs USDk because the canonical input schema uses that unit. It does not imply that the opportunity is located in a dollar economy or that exchange-rate effects have been modelled. A real comparison involving several currencies would need a conversion basis, timing and scenario treatment. A label alone cannot make values commensurate. The same applies to nominal versus real amounts, present values versus undiscounted totals, and payments to the institution versus economic impacts on a broader public perimeter.

The payoff horizon determines what the state comparison can say. An aggregate lifetime present value and a near-term cash inflow are not interchangeable. A committee may care about both, but it should use distinct performance measures or a properly integrated model. The simple app accepts aggregate supplied payoffs and does not discount a dated cash-flow schedule. Its output is an expected monetary comparison of those inputs. If delayed cash, financing pressure or recurrent operating effects dominate the choice, the author should calculate them in an appropriate canonical model or identify the missing analysis.

State payoffs should include the consequences of failure that the institution can actually bear. A failure cell might include an abandoned investment, remediation, service interruption or an exit cost. It might exclude a protected right that cannot lawfully be traded for a larger expected return. These exclusions have different meanings. A cash consequence excluded for lack of evidence creates an uncertainty gap; a non-compensable constraint creates an admissibility boundary. The committee should not resolve the latter by inserting an arbitrary monetary penalty and allowing the optimiser to trade it away.

Success can carry costs too. An action may succeed commercially while imposing operating burden, increasing concentration, creating a liability or consuming capacity needed elsewhere. Calling a cell success should not make it a gross revenue figure. The payoff should represent the relevant net consequence under that state. Where costs remain contingent within the success state, an aggregate may hide variation that matters to risk tolerance. The analyst should disclose that compression and test whether a richer state representation is required. A positive expected payoff does not establish tolerable downside exposure.

The hold baseline deserves the same diligence as an active option. The fixture supplies no monetary consequence for Hold in either represented state. This is a transparent illustrative simplification. In an actual decision, holding may preserve capital while losing a licence, a customer, a delivery slot or a learning opportunity. It may also avoid a cost that is unavoidable under commitment. If those effects are material, a zero baseline would misstate the action comparison. The input owner should explain the baseline rather than treating inaction as an inherently neutral alternative.

The proposed study cost is separated from the action payoffs in the app. This permits the model to display gross information value and then subtract the purchase price once. It also exposes a frequent perimeter error: a team may include the same research cost within an action payoff and subtract it again as the test cost. The paper's fixture convention avoids that double charge by stating that the action payoffs precede the proposed signal purchase. A real commission must review its cost categories so that acquisition, analysis and implementation charges are each included exactly where they belong.

Research costs can have several incidences. A vendor price is only one component. Internal analysts may need time, an operating trial may interrupt service, and a decision may wait while a report is reviewed. The binary engine has a single testCost field. It can accept an aggregate cost if the components have been calculated and documented elsewhere, but it does not estimate those components. A decision package should show how the aggregate was assembled, who bears it, and whether any important nonmonetary consequence is handled as a condition outside the expected-value comparison.

Payoffs are often the more contentious inputs because they express institutional priorities. Different parties may agree on state probabilities and still rank the actions differently when their consequences differ. A sponsor may bear construction cost while an operator bears downtime; a public institution may value continuity while a private party values cash return. The paper does not resolve those differences through a single universal monetary objective. It recommends a declared decision owner and payoff perimeter, with distributional consequences and protected constraints reviewed explicitly. The same evidence can have different value to different legitimate decision makers.

The original completed matrix below gives a functional account of each cell. It does not invent contracts, clients or actual estimated returns. Its purpose is to ask how the supplied input would be supported in a real dossier. For Commit, the review would need the incremental success case and failure exposure. For Stage, it would need an account of the compressed subsequent decisions. For Hold, it would need the opportunity cost and any obligation that persists. For the study cost, it would need a distinct commissioning budget and cost boundary.

Sensitivity should test coherent alternative payoffs, rather than changing isolated numbers without considering their meaning. A reduced upside could reflect a smaller opportunity, a delay already embedded in supplied aggregate values, or a more conservative commercial case. Each interpretation changes the question. The paper's reduced-upside fixture simply supplies a different matrix and discloses that convention. It does not calculate a time discount or infer how long a study takes. Coherent inputs are more important than the attractive appearance of a sensitivity chart.

The committee should also record why it accepts a monetary objective for this particular choice. A risk-neutral expected-value rule can be useful for repeated, diversifiable or limited exposures, but the app cannot prove that these conditions hold. A single concentrated decision can pose liquidity or mandate constraints that an aggregate payoff does not capture. The appropriate response may be a constraint, a utility analysis, a stress test or a different action set. These additions should be specified as additional methods, rather than assumed to emerge from a high expected value.

An acceptable payoff matrix therefore has both a numerical and a narrative audit trail. Numerical checks establish units, arithmetic and consistent cost placement. The narrative trail establishes incidence, timing, baseline, exclusions and reasons for the chosen objective. A reviewer should be able to disagree with a cell while understanding exactly what it means. That is a stronger foundation for a committee debate than an apparently precise matrix whose numbers were agreed only because nobody asked what they excluded.