What does an observed dynamic sublinear valuation rule reveal about latent uncertainty? We establish that, under natural axioms, it encodes latent information about discounting and state-process laws through the geometry of its local valuation mechanism. From the valuation rule, we recover a generating function describing local valuation responses. Its support sets determine the maximal robust family of discounted state-process laws consistent with observed valuations, while its gradient sets determine, for each smooth test payoff, the locally binding models for valuation. Thus observed valuations set-identify not a single hidden model, but exact uncertainty structures about discounting and state evolution. We also show that this information is recoverable from data: even with a restricted payoff class and discrete maturities, one can characterize the identified region of local uncertainty and consistently recover its maximal counterpart. Valuation therefore contains enough information to identify the uncertainty structures.