Detecting Target Leakage by Label Reconstruction: A Diagnostic with a Learning-Analytics Case Study

Data and R code for "Detecting Target Leakage by Label Reconstruction: A Diagnostic with a Learning-Analytics Case Study." The case study uses 118 de-identified peer reviews written by 43 students in an undergraduate visual analytics course (spring 2025). Each review scores a classmate's visualization on an eight-item design rubric and adds free-text comments. The "active reviewer" label (more words than the median review) is a deterministic function of the review's word counts. The paper uses this case to show how such target leakage can be detected.