The prefrontal cortical areas exhibit a wide range of diverse and mixed representations involving sensory, motor, and autonomous variables. Despite this diversity, the prefrontal areas are known to regulate very specific behavioral functions. This disparity between the representations and their precise roles in behavior prompts a fundamental question: which representations do animals deploy, and in what specific behavioral contexts? To address this question, we recorded neurons in the medial prefrontal cortex (mPFC) of mice engaged in probabilistic reward-based foraging task. Using reinforcement learning (RL) model that takes into account both reward and choice history we derived decision variable (DV) that adequately described behavior. We found that neurons integrate choice and reward history in line with their behavioral effects. To probe under what behavioral contingencies DV was used we subjected animals to different behavioral manipulations. These manipulations included altering the conditional dependence of reward probability on past choices and varying the temporal intervals between choices. While neural representations tracked the task contingencies, inactivation of mPFC specifically degraded performance of the animals when there were long temporal gaps between choices. We discovered that the neural representations had to be viewed within the context of animals performance. Namely, we found correlation of animal’s performance and decoding accuracy at the population level only in the task with long temporal gaps between choices. We conclude that if neural representation are examined alone almost identical representations can have different behavioral impacts. Our findings argue that redundant computations exist in medial prefrontal cortex and its behavioral deployment is conditioned on temporal gaps between task relevant events.