Extensive literature in artificial intelligence in education focuses on developing automated methods for detecting cases in which students struggle to master content while working with educational software. Such cases have often been called "wheel-spinning," "unproductive persistence," or "unproductive struggle." We argue that most existing efforts rely on operationalizations and prediction targets that are misaligned to the approaches of real-world instructional systems. We illustrate facets of misalignment using Carnegie Learning's "MATHia" as a case study, raising important questions being addressed by on-going efforts and for future work. [This paper was published in: I. Bitencourt, M. Cukurova, K. Muldner, R. Luckin, & E. Millan (Eds.), "Proceedings of the 21st International Conference on Artificial Intelligence in Education" (AIED 2020). Lecture Notes in Computer Science (LNCS, Vol. 12164 pp.92-97). Springer, Cham.]