Research on the effectiveness of Intelligent Tutoring Systems (ITSs) suggests that automatic hint generation has the best effect on learning outcomes when hints are provided on the level of intermediate steps. However, ITSs for programming tasks face the challenge to decide on the granularity of steps for feedback, since it is not a priori clear which sequence of code edits constitutes a step. We argue that the step granularity for programming tasks should be founded on pedagogical considerations and suggest Vygotsky's theory as an example. Furthermore, we compare several automated methods for sub-dividing programming traces into steps. To evaluate these methods, we provide a novel dataset consisting of 44 code-traces from introductory Python tasks. Furthermore, we provide a novel tool for annotating steps in programming traces and perform a study with six experienced annotators. Our post-survey results show that the annotators favored solved subtasks as a granularity for step division. However, our results show that completed lines as indications of steps explain the annotations best, suggesting that even simple rule-based approaches are suitable for automatic programming trace sub-division. [For the complete proceedings, see ED675583.]