[2603.27076] When Verification Hurts: Asymmetric Effects of Multi-Agent Feedback in Logic Proof Tutoring

Computer Science > Artificial Intelligence arXiv:2603.27076 (cs) [Submitted on 28 Mar 2026] Title:When Verification Hurts: Asymmetric Effects of Multi-Agent Feedback in Logic Proof Tutoring Authors:Tahreem Yasir, Sutapa Dey Tithi, Benyamin Tabarsi, Dmitri Droujkov, Sam Gilson Yasitha Rajapaksha, Xiaoyi Tian, Arun Ramesh, DongKuan (DK)Xu, Tiffany Barnes View a PDF of the paper titled When Verification Hurts: Asymmetric Effects of Multi-Agent Feedback in Logic Proof Tutoring, by Tahreem Yasir and 8 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) are increasingly used for automated tutoring, but their reliability in structured symbolic domains remains unclear. We study step-level feedback for propositional logic proofs, which require precise symbolic reasoning aligned with a learner's current proof state. We introduce a knowledge-graph-grounded benchmark of 516 unique proof states with step-level annotations and difficulty metrics. Unlike prior tutoring evaluations that rely on model self-assessment or binary correctness, our framework enables fine-grained analysis of feedback quality against verified solution paths. We evaluate three role-specialized pipelines with varying solution access: Tutor (partial solution access), Teacher (full derivation access), and Judge (verification of Tutor feedback). Our results reveal a striking asymmetry: verification improves outcomes when upstream feedback is error-prone (<70% accuracy), but degrades perfor...