Reasoning traces have become a valuable form of learning signals for improving and transferring the capabilities of large language models. In particular, detailed traces can help distill reasoning behavior from stronger teacher models into weaker student models. The value of capability transfer has motivated many deployed systems with reasoning models to hide raw internal traces and expose at most summaries and answers to users. As a result, we ask whether such interface-level trace hiding prevents users from obtaining useful reasoning supervision through prompting. We study this question with Reasoning Exposure Prompting (), a lightweight in-context elicitation method that uses shadow-model-generated demonstrations wrapped in auxiliary code-like formats to raise user-visible reasoning traces from a victim model. Across the common reasoning dataset, different victim models, and different student model distillation, substantially increases similarity between exposed and REP-conditioned internal traces while preserving useful reasoning signals.

Grammatical error diagnosis is an essential part in a language-learning tutoring system. Based on the data sets of Chinese grammar error detection tasks, we proposed a system which measures the likelihood of correction candidates generated by deleting or inserting characters or words, moving substrings to different positions, substituting prepositions with other prepositions, or substituting words with their synonyms or similar strings. Sentence likelihood is measured based on the frequencies of substrings from the space-removed version of Google n-grams. The evaluation on the training set shows that Missing-related and Selection-related candidate generation methods have promising performance. Our final system achieved a precision of 30.28% and a recall of 62.85% in the identification level evaluated on the test set.