Assessment of students' programming submissions has been the focus of interest in many studies. Although the final submissions capture the whole program, they often tell very little about how it was developed. In this paper, we are able to look at intermediate programming steps using a unique dataset that captures a series of snapshots showing how students developed their program over time. We assessed each of these intermediate steps and performed a fine-grained concept-based analysis on each step to identify the most common programming paths. Analysis of results showed that most of the students tend to incrementally build the program and improve its correctness. This finding provides us with evidence that intermediate programming steps are important, and need to be taken into account for not only improving user modelling in educational programming systems, but also forproviding better feedback to students.