Micro-Online and Micro-Offline Learning Reflect Stable Yet Randomly Distributed Strategies in Motor Skill Acquisition

Motor learning is traditionally conceptualized as gains emerging during and after practice, over minutes to days. However, learning can also occur on much shorter timescales, within (micro-online) and between (micro-offline) practice blocks over seconds. It remains unclear whether micro-learning reflects an intrinsic property of the learning brain, arises from mechanisms such as reactive inhibition, or instead reflects transient fluctuations, stable individual traits, or chance. We pooled data from five motor sequence learning experiments (146 participants of either sex), including healthy younger and older adults and patients with Parkinson's disease (PD). Participants practiced motor sequences across repeated blocks separated by short rests. Using linear mixed-effects models, we quantified micro-online and micro-offline learning, related them to macro-learning, and assessed effects of age, PD, and dopaminergic medication. Micro-online learning predominated early in training, while micro-offline learning became the primary driver of later gains; neither modality alone predicted macro-learning. Individual preferences for one modality were stable across sessions, indicating trait-like profiles. However, the strong negative correlation typically reported between the two modalities was largely explained by how they are mathematically derived, rather than a genuine behavioral trade-off. Older participants relied more heavily on micro-online learning in early stages and showed stronger modality preferences, despite comparable total learning to younger adults. Patients with PD off medication showed reduced micro-online but enhanced micro-offline learning, a shift not observed under dopaminergic medication. However, group-level effects were modest relative to pronounced interindividual variability, and the observed micro-learning patterns may partly reflect measurement properties rather than distinct learning processes. Significance Statement Motor learning is typically assessed over minutes to days, yet performance also changes over seconds, within (micro-online) and between (micro-offline) brief practice bouts. Whether these rapid dynamics reflect genuine learning, transient effects such as reactive inhibition, stable individual differences, or chance has remained unclear. Here, we show that micro-online and micro-offline learning reflect interacting processes whose relative balance varies widely across individuals yet remains stable within individuals, forming trait-like profiles shaped selectively, rather than broadly, by age and Parkinson's disease. We further show that the commonly reported trade-off between the two modalities partly reflects how they are mathematically computed rather than genuine competition. Critically, no single micro-learning profile was linked to poorer outcomes, indicating multiple equivalent pathways to successful learning.