Decision-makers in high-stakes, high-stress, and time-limited scenarios (such as emergency first-responders) face challenges in factoring in all relevant information, making optimal choices as a response, and avoiding choice paralysis when decisions are difficult or demand ethical considerations. This work attempts to better advise human deciders in these systems by introducing and formalizing such Constrained Causal Decision Dilemmas (C2D2s) as sequential decision-making scenarios characterized by (1) decisions that differently affect the environment vs. the agent’s perspective of the environment at different tiers of the Pearlian causal hierarchy, (2) restrictions in choices imposed by some conditions like a time limit, and (3) some possibly multi-dimensional utility describing multiple objectives desired in an outcome. To provide optimal recommendations in C2D2s, we propose a Causal Decision Network (CDN) used in companion with a novel type of Causal Expectimax Search (CES) that can plan for the best Course of Action (COA). Not only do simulations demonstrate the efficacy of these techniques compared to traditional, associative approaches, but also expose a number of interesting edge cases to causal decision-making that are handled by CDNs and CES: (a) the Ostrich Effect, by which agents may avoid investigating some variable should doing so reduce expected utility, (b) the Possum Effect, during which the best choice may be to make no choice at all, (c) the Double Check Effect, in which the best policy is to investigate the outcome of an intervention and repeat if the result is unsatisfactory, and (d) the Pseudo-counterfactual Effect, by which an optimal choice may involve changing the result of an investigation through intervention. Taken together, the contributions of this work include the formalization of a new class of decision problems, demonstration of causal tools in their solution, and discussion of applications to data-sparse domains.