Abstract During decision making, choices are made based on assessing potential options and their expected outcomes. Traditional laboratory investigations of decision making often employ tasks involving the discrimination of perceptual evidence, where sensory information is constant and presented continuously. However, during natural behavior, this is unlikely the case. Usually, perceptual information is dynamic and presented intermittently, which requires maintaining information in memory. Thus, understanding decision making requires considering the dynamics of working memory. Here, we used a perceptual decision-making task where fifteen tokens jump from a central circle to one of two peripheral ones and disappear shortly after. Participants were required to report which target they believed would have received most tokens by the trial’s end. Half of the trials included a temporal gap, during which no information was displayed. In those cases, we found that participants made choices with less available information, but their accuracy remained unchanged. Computational modeling revealed that this behavior was best explained by a model in which stored perceptual information leaks away due to the arrival of new information, rather than by the passage of time. Our results provide evidence of a decision-making process that evolves even in the absence of perceptual information, challenging the idea of a frozen state resilient to temporal gaps and shedding light on the dynamics of working memory. This study highlights the importance of considering working memory dynamics in understanding decision-making processes, particularly in environments with intermittent perceptual information. Significance statement Our research challenges the notion of a decision-making process that freezes in the absence of perceptual information. Through a novel task with temporal gaps, we demonstrate that decision making continues to evolve even when perceptual cues are absent. Additionally, we highlight the importance of working memory dynamics in such process. We show that choices are the result of a combination of mnemonic evidence with urgency, a signal that reflects the need to respond. Computational modeling supports a working memory model where stored perceptual information leaks away due to the arrival of new events but remains stable between events. These findings offer insights into the decision-making process, emphasizing the importance of considering working memory dynamics in understanding human behavior.