Detecting Answer-Driven Reasoning in LLM-Based Educational Tutors via Truncated Chain-of-Thought Auditing
2026-07-06 • Artificial Intelligence
Artificial IntelligenceMachine Learning
AI summaryⓘ
The authors studied how language models used as math tutors might produce answers based on hidden correct or incorrect solutions, rather than fully working through the problem. They tested whether the model’s explanation shows the answer too early, using a method called TRACE that checks how soon an answer can be predicted from the explanation. They found that when the model has access to the correct answer, it often reveals the answer very early in the explanation, even before fully justifying it. This suggests a way to check if a tutor model is truly reasoning through problems or just leaking answers.
Large language modelChain-of-thought reasoningTruncated Reasoning AUC Evaluation (TRACE)Math tutoring systemsAnswer leakageGSM8K datasetModel explanation auditingAnswer key accessBehavioral evaluation
Authors
Bonan Shen, Dingyan Shang, Youting Wang, Tao Ning
Abstract
Large language model (LLM) tutors often produce fluent step-by-step explanations, but a correct and pedagogically formatted response does not guarantee that the answer was derived from the student-facing problem. In realistic tutoring systems, the model may also have access to teacher notes, answer keys, rubrics, or retrieved solution artifacts. We study whether such private answer information can make tutor explanations answer-driven: the final answer is behaviorally available before the written explanation has justified it. Using Truncated Reasoning AUC Evaluation (TRACE), which probes how early a chain-of-thought prefix can pass a verifier, we evaluate 1000 GSM8K test problems under three paired tutoring contexts: question-only, correct answer-key, and wrong answer-key. At fixed fractions of each generated explanation, we force the model to answer immediately and verify the response against the gold numeric answer. With Qwen2.5-3B-Instruct, answer-key access raises median TRACE AUC from 0.375 to 0.900 and makes the gold answer available at the first 10% prefix in 997 of 1000 cases. The effect remains strong on the 746 examples where both question-only and answer-key explanations end with the correct answer. These results support truncated CoT auditing as a lightweight process-level diagnostic for answer-driven reasoning in math tutoring explanations.