ABSTRACT Imagine baking a croquembouche, a new culinary challenge. How many steps ahead in the recipe will you bother to remember? Of course, you can always refer back to the recipe itself, so you needn’t remember much. Everyday tasks involve strategic trade-offs between sampling (utilizing the external resources provided by stable objects and symbols in the visual world) versus remembering (utilizing the internal resources of working memory). Here our goal is to answer the question: When it’s up to you, how much will you remember? Historically, working memory has been measured under ideal conditions – well-informed, motivated participants doing simple tasks – to isolate best-case performance. Estimates of WM from these types of studies, however, provide a challenge for generalization: real-world conditions are not ideal and performance is rarely best-case. Our view, captured by our dynamic feedback model, is that an individual trades off sampling versus remembering in order to strike a preferred balance between the subjective costs of using external resources (the overhead of having to refer to information in the environment) with the subjective costs of using internal resources (the cognitive effort required to maintain task-relevant information in WM). Nearly all of the previous empirical work investigating the sampling-remembering trade-off has been with adults. This leaves a tremendous gap in our understanding since the optimal use of internal resources such as WM is even more crucial when those resources are more limited, as in children. We developed an innovative “Shopping Game”, a child-friendly, tablet-based paradigm where children are shown a shopping list, then asked to select those items from a virtual store. Since the list is long, and the list and store are not visible at the same time, the child must toggle back and forth between the two. Through the parametric variation of time delays to access the store or the list, we will independently manipulate the subjective costs to use those internal and external resources, and through pupillometry, we will track moment-to-moment cognitive effort deployment. Importantly, we propose to study these mechanisms in a large longitudinal study of 6-7-year-old children, during a period right after school entry, when children are first faced with cognitive tasks that require memorization. Children’s steadily increasing WM capacity during this period provides a unique opportunity to test predictions of our model that could not be tested in adults. Combining psychophysics, pupillometry, and standardized cognitive assessments, we will conduct a critical multi-method, naturalistic test of how the cognitive control mechanisms of working memory use develop. Our project will provide (1) the first model-based analysis of the sampling-remembering trade-off and (2) the first direct, longitudinal study of the deployment of cognitive effort in children.