Extended Data Fig. 9: Additional analyses of the rodent dataset.

For the rodent data recorded by Widloski and Foster⁷, we quantified the shortest initial distance-to-goal on each trial as well as (i) the time spent at the previous well before initiating the trial, and (ii) the number of replays detected during this period. (a) Pearson correlation during home trials between initial distance-to-goal and either (i) time spent at the previous well (Δt, left), or (ii) the number of replays performed at the previous well (right). Bars and error bars indicate mean and standard error across sessions (gray dots; n = 37). The absence of a correlation between initial goal distance and time spent at the previous well differs from our analyses of human behavior in a similar maze task (Fig. 2c). However, there are two notable differences between these two paradigms that might explain the apparently discrepant results. Firstly, the rats recorded by Widloski and Foster⁷ have to physically consume the reward at the previous well before they can continue their behavior. Secondly, there is an experimenter-imposed delay between the end of reward consumption and the next reward becoming available. This is different from the human task paradigm, which was explicitly designed to encourage a trade-off between the time spent thinking and the time spent acting, without any additional ‘down time’ that could be used for planning without incurring a temporal opportunity cost. (b) As in (a), now for away trials. (c) Fraction of replays reaching either the true goal (left) or a randomly sampled alternative goal location (right) during away trials. Dashed lines indicate individual sessions (n = 37), and solid lines indicate mean and standard error across sessions. In contrast to the home trials (Fig. 4c), the goal is not over-represented during away trials, where the goal location is unknown. (d) Over-representation of replay success as a function of replay number within sequences of replays containing at least 3 distinct replay events (c.f. Fig. 4e). Bars and error bars indicate mean and standard error across replays pooled from all animals. In contrast to the home trials, there is no increase in over-representation with replay number during these away trials.

Source data