Abstract: An inherent feature of learning in most disciplines, especially STEM, is its cumulative structure, which makes developing advanced skills challenging. This paper credibly estimates a dynamic learning technology of student effort in a foundational university course with a cumulative structure. Doing so is incredibly difficult because effort inputs are typically unobserved and are dynamic endogenous choices. To address this, I use rich panel data on nearly 3,700 undergraduates in an online introductory programming course, which precisely tracks study time throughout the course. Then I carry out a field experiment which generates period-by-period exogenous variation in effort allocation, enabling me to identify dynamic interactions across effort inputs in the learning technology. I find evidence of dynamic learning complementarities as the marginal benefit to studying in each learning period is increasing in prior knowledge accumulated. I then develop and estimate a multi-stage behavioral model of effort supply, using the experiment to identify benefit and cost parameters at each stage. The model informs policy simulations of grading schemes, showing that decreasing assignment weights across the course—rather than equal weights—boosts effort early on when foundational skills are acquired. The findings in this study have implications for effective learning strategies and approaches to course design