["British Journal of Educational Technology, EarlyView. ", "\n\nAbstract\nSelf‐regulated learning (SRL) is an active process involving multiple interacting components that evolve over time, exhibiting characteristics of complex systems such as non‐linearity, emergent behaviour, self‐organization, and hierarchy. These interactions unfold at different temporal levels, each warranting a dedicated lens to capture their distinct dynamics. In this study, we apply a complex dynamic systems lens to analyse the longitudinal dynamics of SRL. We map how different SRL processes interact with each other across time and scales: (1) the stable between‐person level, which represents the dominant approach to learning or roughly the trait of SRL, (2) the contemporaneous level, which maps how SRL processes influence each other within the same time and (3) the temporal level, which captures how processes predict or influence each other in the future. Data were collected through a weekly survey administered over 4 weeks in five courses at two institutions, complemented by LMS behavioural engagement data. A panel vector autoregression model was employed to examine the structure and dynamics of SRL and LMS behavioural engagement at the three levels. The findings suggest that central SRL processes, such as planning and adapting, take place in separate stages, in accordance with the classic SRL models, whereas other processes, like effort regulation, are more pervasive, co‐occurring with most other regulatory processes. At the aggregate level, adjusting was the most central process that drove students' SRL. As such, our results align with the main characteristics of complex systems, including non‐linearity and hierarchy. These findings have implications for the design of SRL interventions, where effort can benefit from real‐time prompts, whereas metacognitive processes might require long‐term scaffolding. Furthermore, the weak association between LMS engagement and SRL processes across all levels highlights the limitations of relying solely on behavioural trace data to infer regulation.\n\n\n\n\nPractitioner notes\nWhat is already known about this topic?\n\nSelf‐regulated learning (SRL) is an important driver of academic success and can be influenced through targeted interventions.\nMost SRL research is based on group‐level data, often using static, cross‐sectional designs that overlook temporal dynamics.\nRecent work has highlighted that SRL can and should be modelled as a complex dynamic system.\n\nWhat this paper adds?\n\nThere is evidence of complex systems characteristics in SRL such as hierarchy, non‐linearity and feedback loops.\nSRL processes follow distinct temporal phases, with some processes persisting throughout all phases.\nMetacognition is the most central process at the between‐person level, whereas effort is central at the within‐person level.\nLMS behavioural data is weakly linked to self‐reported SRL.\n\nImplications for practice and/or policy\n\nSRL interventions should consider how regulatory processes unfold over time, rather than treating SRL as a static trait.\nInterventions targeting effort regulation and metacognition have the potential to be the most consequential.\nCaution must be exerted when using average or between‐person data to inform individualized support.\nLMS metrics should be interpreted with care and ideally complemented by self‐report or observational data.\n\n\n\n\n"]