I study a preference relation on risky long run public projects induced by a large maturity limit of expected present values. Under common assumptions this relation has a variational representation that is related to a well known model of ambiguity aversion; it is non-probabilistic in general. The formalism generalizes Weitzman's `lowest possible rate' formula for long run discount rates to a large class of stochastic economies, gives rise to a notion of stochastic dominance adapted to long run valuation, and characterizes features of stochastic processes that cause long run cost benefit rules to be non-probabilistic.