STR.9:4.2 - Choose the criterion that answers the actual decision

State what the chooser is trying to protect or improve. Expected-value comparison can be useful with defensible probabilities and a suitable value model. Under deep uncertainty, conditional performance, worst-case loss, regret or an acceptable region may serve better.

These criteria answer different questions. Worst-case comparison emphasizes the largest loss in the declared conditions. Regret compares an option’s result with the best available alternative in each condition, under a common justified measure. A small regret does not make a breach of a protected obligation acceptable.

Keep material value disagreement explicit. A scalar can help only when its trade-offs are warranted for this use. Do not invent equal scenario probabilities, silently normalize unlike consequences or change the criterion after seeing the winner.

For a dominance claim, specify the compared characteristics, their preferred directions, conditions and evidence. Being no worse on every relevant coordinate and better on at least one can support dominance under that rule. A weighted-score win or a higher average does not establish dominance. Where a required consequence estimate is missing, leave the coordinate comparison unresolved. Non-dominated candidates still require a selection judgement.