The paper models an individual who may not foresee all relevant
aspects of an uncertain environment. The model is axiomatic and
provides a novel choice-theoretic characterization of the subalgebra
of foreseen events. It is proved that all recursive, consequentialist
models imply perfect foresight and thus cannot accommodate unforeseen contingencies. In particular, the model is observationally distinct from recursive models of ambiguity. The process of learning implied by dynamic behavior generalizes the Bayesian model and permits the subalgebra of foreseen events to expand over time.