C.43:4.4 - Compare whole continuations from the actual starting conditions

Make alternatives answer the same question. Align the starting conditions, time horizon, outcome descriptions and resource assumptions. Retain differences that matter to affected participants. Apply C.11.CRC to construct what a finite change adds relative to the current configuration, and C.11 to select among the warranted options.

A positive contribution relative to the present configuration is not yet a choice among available continuations. Another attainable use of the same means may be better. Conversely, a locally attractive component can fail when it requires unavailable complements, conflicts with another action or removes a later correction. Compare the connected consequences, including transition work, interactions and remaining choices.

Value information through the decisions it can improve, together with any separately justified use of the knowledge. For one decision maker with a specified probability model, a fixed finite set of available actions A, integrable utility u, and current information I, the gross instrumental value of a signal Y can be written as:

[ \mathbb E_Y!\left[\max_{a\in A}\mathbb E[u(a,\theta)\mid I,Y]\right] -\max_{a\in A}\mathbb E[u(a,\theta)\mid I]. ]

The expression compares choosing after the signal with choosing without it. It assumes the signal can be ignored and the former actions retained. Acquisition, interpretation, delay and displaced opportunities enter the net comparison. If the experiment changes the system or available actions, compare those changed continuations rather than subtracting a price from an otherwise unchanged formula.

The existing information matters. Two signals that are useful separately can repeat one another; a weaker isolated signal can be more useful alongside what is already known. Brooks, Frankel and Kamenica’s comparison of signals provides the source basis for this dependence. Entropy reduction is not by itself a substitute for the receiving decision’s gain.

Keep feasibility, outcome preference and assurance distinct in the comparison. An option can be preferred but unavailable; an admissible option can be inferior; a feasible finite plan can leave no acceptable continuation. Use the strength of calculation or evidence needed for the actual conclusion. A local estimate supports a local choice under its assumptions; a safety guarantee or optimality claim requires the relevant stronger construction.