While Large Language Models (LLMs) excel in various general domains, they exhibit notable gaps in the highly specialized, knowledge-intensive, and legally regulated Chinese tax domain. Consequently, while tax-related benchmarks are gaining attention, many focus on isolated NLP tasks, neglecting real-world practical capabilities. To address this issue, we introduce TaxPraBen, the first dedicated benchmark for Chinese taxation practice. It combines 10 traditional application tasks, along with 3 pioneering real-world scenarios: tax risk prevention, tax inspection analysis, and tax strategy planning, sourced from 14 datasets totaling 7.3K instances. TaxPraBen features a scalable structured evaluation paradigm designed through process of "structured parsing—field alignment extraction—numerical and textual matching", enabling end-to-end tax practice assessment while being extensible to other domains. We evaluate 19 LLMs based on Bloom’s taxonomy. The results indicate significant performance disparities: all closed-source large-parameter LLMs excel, and Chinese LLMs like Qwen2.5 generally exceed multilingual LLMs, while the YaYi2 LLM, fine-tuned with some tax data, shows only limited improvement. TaxPraBen[<https://anonymous.4open.science/r/TaxPraBen/>] serves as a vital resource for advancing evaluations of LLMs in practical applications.

Large Language Models (LLMs) achieve strong performance through extended inference-time deliberation, yet how their reasoning failures arise remains poorly understood. By analyzing model-generated reasoning trajectories, we find that errors are not uniformly distributed but often originate from a small number of early transition points, after which reasoning remains locally coherent but globally incorrect. These transitions coincide with localized spikes in token-level entropy, and alternative continuations from the same intermediate state can still lead to correct solutions. Based on these observations, we introduce GUARD, a targeted inference-time framework that probes and redirects critical transitions using uncertainty signals. Empirical evaluations across multiple benchmarks confirm that interventions guided by these failure dynamics lead to more reliable reasoning outcomes. Our findings highlight the importance of understanding when and how reasoning first deviates, complementing existing approaches that focus on scaling inference-time computation.

"The 24th Chinese Computational Linguistics Conference (CCL25-Eval) features 12 technical evaluation tasks. Among them, Task 7 is the Chinese Literary Language Understanding Evaluation (ZhengMing). ZhengMing is a universal and scalable evaluation framework designed to assess natural language processing (NLP) tasks in the literary domain, such as text classification, text generation, automated question answering, relation extraction, and machine translation.ZhengMing framework aims to evaluate the performance of large language models (LLMs) in the literary field at a fine-grained level. In this mission, 89 teams signed up for the competition, with5 teams ultimately submitting results. The highest score achieved is 0.65. This paper presents and discusses the dataset, task descriptions, competition results, and other relevant information for this evaluation task. This paper introduces and presents relevant information about this evaluation task, including the dataset, task description, and competition results. More details are available at https://github.com/isShayulajiao/CCL25-Eval-ZhengMing."