Outcome lottery

A distribution over outcomes; expectation is one functional among many.

Role. Target. Graph node dist.

Expected value, CVaR, stochastic dominance, entropic risk, chance constraints, minimax, and regret can rank the same lotteries differently.

— Frames draft §4.2 (Bäuerle and Jaśkiewicz 2024)

Distributional RL learns a random return and still leaves open which functional governs action. Projecting to a mean state score has no section: the tails and model ambiguity are gone.