Savage Without Monotonicity

Savage separates belief from taste: coherent preferences over acts can be represented as subjective probabilities over states together with utilities over consequences. We show a boundary of that separation. Savage is the totalizing form of this ideal: a two-sided universal domain, covering both state-contingent acts and events, whose representation delivers a finitely additive probability together with fine-grained infinite uncertainty. That package can supply a positive event A decomposed into disjoint null cells E_n; we isolate both a countable convex-range route and a constructibility-based null-ideal route to that partition. On a real-valued outcome scale, dominance naturally supplies strictly better consequences whose utility gains tend to zero; expected utility can then tie an act with its statewise improvement. Dominance says the improved act should be strictly preferred, because the improvement occurs throughout a behaviorally non-null event. Representation says the two acts are indifferent. The result is conditional: countable additivity, an explicit dominance axiom, an act-domain restriction, a consequence restriction, or a state-space restriction blocks the construction. Its philosophical point is that disentangling belief and taste does not itself supply monotonicity. Trustworthy ideal-decision arguments need to say which extra principle carries the dominance work.