The Correct Answer Trap: Pedagogically-Grounded Detection and Feedback for Hidden Misconceptions

The study presents a novel approach to detecting hidden misconceptions in student responses using 20,964 samples from the Eedi mathematics platform. While fine-tuned classifiers achieve a 57% detection rate for these misconceptions, an open-weight reasoning model improves this to 84%, although it suffers from a high false alarm rate of 8 to 1. The proposed detect-verify-escalate pipeline aims to enhance feedback mechanisms by separating correctness from method validity and adapting deployment for both teacher dashboards and autonomous tutoring systems, offering a more nuanced response to student misunderstandings.