Xu, H.; Wen, W.; Yu, Y.; Mehmood, I.; Dai, Z.; Mariam, M.; Zhang, B.; Yu, X.

Predictive processing theories posit that the auditory system continuously generates expectations and updates internal models when incoming evidence violates those expectations. Whether such updating can be expressed as a temporally precise, change-like neural response at a context-defined moment during continuous stimulation, without any physical change, has remained unclear. Using a transitional click-train paradigm in human EEG, we show that listeners reliably detect subtle midpoint transitions when click timing is fixed, whereas temporal uncertainty markedly weakens this sensory-driven detection. Strikingly, even in a physically unchanged stimulus, variable click timing elicits a robust response time-locked to the nominal midpoint. This endogenous change-like signal is stimulus-statistics dependent, absent in the fixed-timing no-change control, temporally confined to the early post-change window, and strongly state dependent, being amplified during active change detection relative to passive listening. Moreover, within identical no-change trials, the response tracks subjective change reports. Together, these findings show that temporal uncertainty and task engagement jointly gate predictive updating in continuous auditory streams, providing a compact assay to dissociate sensory-driven change responses from context- and state-dependent predictive signals.