包裹，长文用
/
划分小节；列表用
- ，引用用
，代码用
，重要句子用；保留所有原文信息，不得增删改；不要输出文档级标签。
接下来，我需要分析用户提供的正文内容。正文是一段关于GUIDER方法的学术论文摘要，内容涉及临床实践指南、推理模型训练、现有方法的局限性、GUIDER的设计与优势，以及实验结果。整段内容是一个连续的文本，没有明显的小节划分，但需要根据内容逻辑判断是否需要用或
。不过原文本身没有分节标题，所以可能需要直接处理为段落。
然后，我要检查是否有需要用标注的重要句子。根据规则，重要句子可以用，但用户没有明确哪些是重要的，可能需要根据内容的关键点来判断。例如，GUIDER方法的提出、其解决的两个问题（path-answer gap和path multiplicity）、实验结果等可能是重点。
现在，逐句处理正文。原文是一个长段落，需要拆分成自然的段落吗？原文本身是一个连贯的段落，但可能需要根据逻辑断句。不过用户要求每个自然段落用包裹，所以如果原文是一个自然段落，就用一个
包裹。但原文内容较多，可能需要拆分成几个段落？需要仔细阅读原文结构。 原文结构：首先介绍临床实践指南作为监督来源的难点（path-answer gap和path multiplicity）；然后讲现有方法的局限性；接着提出GUIDER方法及其解决上述问题的方式；最后是实验结果。可以考虑将这段长文本拆分为几个
，但需要确保不改变原意。例如，第一句到“path multiplicity”为一个段落，然后现有方法为一个段落，GUIDER的提出为一个段落，实验结果为一个段落。但需要确认原文是否有自然段的划分，用户可能希望保持原文的自然段落结构，所以如果原文是一个自然段，就只用一个
，但如果内容较长，可能需要拆分。不过原文在用户提供的文本中是一个连续的段落，所以可能需要用一个
包裹，但里面可以用标注关键部分。 另外，检查是否有列表内容，原文中没有列表，所以不需要
。引用和代码也没有，所以不需要
和
。接下来，确定哪些句子需要用。例如，“We introduce GUIDER...”和“GUIDER with Qwen3-14B achieves...”等句子是核心内容，可能需要加粗。
现在，开始构建HTML片段。首先，将整个正文内容放入标签中，然后在关键句子外层加上。例如，“We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.”这句话是方法的提出，是重点，所以加粗。同样，“GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence...”是实验结果，也加粗。其他部分如问题定义“path-answer gap”和“path multiplicity”也可能是重点，可以加粗。
需要注意的是，必须保留原文所有信息，不得增删改内容。所以在添加时，不能改变原文的词语或句子结构，只能在原有句子外层包裹标签。例如，原文中的“which we call the path--answer gap”和“which we call path multiplicity”可以分别在“path--answer gap”和“path multiplicity”外层加，或者整个分句加粗，但需要准确。 现在，逐句处理： 1. “Clinical practice guidelines encode evidence-based recommendations as conditional decision flows, which makes them a natural source of process supervision for training reasoning models.” 这句是背景介绍，可能不需要加粗。 2. “Two properties of clinical reasoning make this supervision difficult to use.” 总起句，不需要。 3. “A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path--answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity.” 这里有两个概念定义，可能在“path--answer gap”和“path multiplicity”外层加，或者整个分句。例如，将“which we call the path--answer gap”和“which we call path multiplicity”中的核心名词加粗。 4. “Current training approaches address parts of this problem.” 总起，不需要。 5. “Answer rewards supervise the final decision, and process rewards computed against a reference flow supervise one valid decision flow per case.” 现有方法，不需要。 6. “Supervised fine-tuning on annotated flows teaches guideline-consistent reasoning but constrains the model to one trajectory per case and can reduce answer accuracy.” 现有方法的局限，可能需要加粗“constrains the model to one trajectory per case and can reduce answer accuracy”部分，但需要看是否重要。 7. “We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.” 方法提出，重点，加粗整个句子。 8. “The controller learns to generate plausible decision flows from guideline-derived supervision, and the policy learns to generate a decision flow and a final answer for each patient case.” 方法细节，可能不需要。 9. “GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the path--answer gap.” 方法优势，解决两个问题，可能需要加粗“which accounts for path multiplicity”和“which addresses the path--answer gap”。 10. “Joint training allows the process supervision to evolve with the policy.” 方法特点，不需要。 11. “On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline.” 实验结果，重点，加粗整个句子。 12. “With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.” 其他实验结果，加粗整个句子。 现在，将这些加粗的部分准确包裹标签，确保不改变原文内容。例如： - “We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.” → We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy. - “GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline.” → On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline. - “With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.” → With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE. 另外，关于“path--answer gap”和“path multiplicity”，可以分别在各自的定义部分加粗，例如： “A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path--answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity.” 同样，“which accounts for path multiplicity”和“which addresses the path--answer gap”中的核心部分也可以加粗，但需要确保准确。例如： “GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the
，代码用，重要句子用；保留所有原文信息，不得增删改；不要输出文档级标签。 接下来，我需要分析用户提供的正文内容。正文是一段关于GUIDER方法的学术论文摘要，内容涉及临床实践指南、推理模型训练、现有方法的局限性、GUIDER的设计与优势，以及实验结果。整段内容是一个连续的文本，没有明显的小节划分，但需要根据内容逻辑判断是否需要用或
。不过原文本身没有分节标题，所以可能需要直接处理为段落。 然后，我要检查是否有需要用标注的重要句子。根据规则，重要句子可以用，但用户没有明确哪些是重要的，可能需要根据内容的关键点来判断。例如，GUIDER方法的提出、其解决的两个问题（path-answer gap和path multiplicity）、实验结果等可能是重点。 现在，逐句处理正文。原文是一个长段落，需要拆分成自然的段落吗？原文本身是一个连贯的段落，但可能需要根据逻辑断句。不过用户要求每个自然段落用
包裹，所以如果原文是一个自然段落，就用一个
包裹。但原文内容较多，可能需要拆分成几个段落？需要仔细阅读原文结构。 原文结构：首先介绍临床实践指南作为监督来源的难点（path-answer gap和path multiplicity）；然后讲现有方法的局限性；接着提出GUIDER方法及其解决上述问题的方式；最后是实验结果。可以考虑将这段长文本拆分为几个
，但需要确保不改变原意。例如，第一句到“path multiplicity”为一个段落，然后现有方法为一个段落，GUIDER的提出为一个段落，实验结果为一个段落。但需要确认原文是否有自然段的划分，用户可能希望保持原文的自然段落结构，所以如果原文是一个自然段，就只用一个
，但如果内容较长，可能需要拆分。不过原文在用户提供的文本中是一个连续的段落，所以可能需要用一个
包裹，但里面可以用标注关键部分。 另外，检查是否有列表内容，原文中没有列表，所以不需要
。引用和代码也没有，所以不需要
和。接下来，确定哪些句子需要用。例如，“We introduce GUIDER...”和“GUIDER with Qwen3-14B achieves...”等句子是核心内容，可能需要加粗。 现在，开始构建HTML片段。首先，将整个正文内容放入标签中，然后在关键句子外层加上。例如，“We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.”这句话是方法的提出，是重点，所以加粗。同样，“GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence...”是实验结果，也加粗。其他部分如问题定义“path-answer gap”和“path multiplicity”也可能是重点，可以加粗。 需要注意的是，必须保留原文所有信息，不得增删改内容。所以在添加时，不能改变原文的词语或句子结构，只能在原有句子外层包裹标签。例如，原文中的“which we call the path--answer gap”和“which we call path multiplicity”可以分别在“path--answer gap”和“path multiplicity”外层加，或者整个分句加粗，但需要准确。 现在，逐句处理： 1. “Clinical practice guidelines encode evidence-based recommendations as conditional decision flows, which makes them a natural source of process supervision for training reasoning models.” 这句是背景介绍，可能不需要加粗。 2. “Two properties of clinical reasoning make this supervision difficult to use.” 总起句，不需要。 3. “A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path--answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity.” 这里有两个概念定义，可能在“path--answer gap”和“path multiplicity”外层加，或者整个分句。例如，将“which we call the path--answer gap”和“which we call path multiplicity”中的核心名词加粗。 4. “Current training approaches address parts of this problem.” 总起，不需要。 5. “Answer rewards supervise the final decision, and process rewards computed against a reference flow supervise one valid decision flow per case.” 现有方法，不需要。 6. “Supervised fine-tuning on annotated flows teaches guideline-consistent reasoning but constrains the model to one trajectory per case and can reduce answer accuracy.” 现有方法的局限，可能需要加粗“constrains the model to one trajectory per case and can reduce answer accuracy”部分，但需要看是否重要。 7. “We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.” 方法提出，重点，加粗整个句子。 8. “The controller learns to generate plausible decision flows from guideline-derived supervision, and the policy learns to generate a decision flow and a final answer for each patient case.” 方法细节，可能不需要。 9. “GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the path--answer gap.” 方法优势，解决两个问题，可能需要加粗“which accounts for path multiplicity”和“which addresses the path--answer gap”。 10. “Joint training allows the process supervision to evolve with the policy.” 方法特点，不需要。 11. “On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline.” 实验结果，重点，加粗整个句子。 12. “With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.” 其他实验结果，加粗整个句子。 现在，将这些加粗的部分准确包裹标签，确保不改变原文内容。例如： - “We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy.” → We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy. - “GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline.” → On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline. - “With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.” → With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE. 另外，关于“path--answer gap”和“path multiplicity”，可以分别在各自的定义部分加粗，例如： “A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path--answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity.” 同样，“which accounts for path multiplicity”和“which addresses the path--answer gap”中的核心部分也可以加粗，但需要确保准确。例如： “GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the