In natural environments, auditory spatial cues are often subject to stochastic variations. Incorporating preceding sensory information as priors makes perceptual spatial inference more robust. However, abrupt changes in the environment lead to surprise and reduce the relevance of those priors, necessitating a rapid establishment of new priors. Thus, in a dynamic environment, quick adaptation of integration time scales is crucial to appropriately balance robustness against flexibility. We study this adaptivity in human listeners using a dynamic auditory localization task with sequences of sounds. First, we contrast conditional Bayesian integration of prior and evidence with a more heuristic approach that momentarily switches between prior and evidence. The best fitting model algorithm is then used to unravel its implementation as measured via high-density EEG. Preliminary results indicate a Bayesian sequential cortical evaluation of evidence, surprisal, and prior updating in response to every single sound in a sequence.