`包裹，长文用`
`/`
`划分小节。 3. 列表内容用`
- `，引用用`
`，代码用`
`，重要句子可用``。
4. 保留原文所有信息，不得增删改内容。
5. 不要输出``/``/``等文档级标签。
正文分析：
正文是一段比较长的英文段落，讨论的是多轮LLM智能体训练中的步进式组相对训练（stepwise group-relative training）及其带来的问题，以及提出的OMA方法。
可以分为几个小节，使其更符合“长文用``/`
`划分小节”的要求。
原文拆解：
1. To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion creates an outer-accounting problem distinct from within-trajectory credit assignment: row-based outcome statistics can count one episode outcome multiple times, while token-level reduction can assign unequal aggregate objective mass across trajectories. We term these effects outcome-statistics multiplicity and objective-mass imbalance.
2. We introduce Outcome-Mass Alignment (OMA), a two-component method that makes both accounting choices explicit. Outcome Alignment computes outcome statistics over unique source trajectories, while Mass Alignment controls aggregate objective mass after within-trajectory aggregation of optimizer-native token objectives. Neither redesigns the base optimizer's credit assignment, importance ratios, or clipping.
3. Across matched ALFWorld GRPO comparisons at two model scales, where failed trajectories typically expand much more than successful ones, OMA improves matched-horizon endpoint success by 7.0-9.4 percentage points. Ordinary WebShop exhibits weak or reversed expansion-outcome coupling and correspondingly mixed effects, whereas in a success-oriented WebShop setting, WebShop-SOL, OMA improves success nAUC from 0.399 to 0.511. Component ablations and GiGPO and GSPO comparisons further support component separability and cross-optimizer compatibility. These results show that trajectory expansion always defines an implicit weighting rule, but its practical effect depends on how expansion couples to outcomes and optimizer-native updates.
划分小节：
可以加两个h2或h3。
问题背景与挑战
-> 但规则说“保留原文所有信息，不得增删改内容”。如果加小标题，算不算增改内容？
通常加HTML结构标签不算增改文本内容，但加中文小标题会引入原文没有的文本。为了严格遵守“不得增删改内容”，小标题应该用原文中的词，或者干脆不加小标题，只用``，或者把原文的第一句或某句作为小标题？
或者，因为原文只有一段，直接用一个`
`包裹即可，或者为了体现“长文用h2/h3划分”，我可以用原文的句子作为小标题？
比如：
Outer-Accounting Problem
(但原文是 "outer-accounting problem"，如果直接用原文的句子或短语，可能会被认为增改。)
更安全的做法：不添加原文没有的文本。如果不需要划分，就只用``。但规则说“长文用`
`/`
`划分小节”，这可以是一个建议（如果需要的话）。原文其实是一段完整的摘要。如果非要划分，可以把原文拆成几个`
`，或者用原文的词做小标题。
比如：
`，代码用``，重要句子可用``。 4. 保留原文所有信息，不得增删改内容。 5. 不要输出``/``/``等文档级标签。 正文分析： 正文是一段比较长的英文段落，讨论的是多轮LLM智能体训练中的步进式组相对训练（stepwise group-relative training）及其带来的问题，以及提出的OMA方法。 可以分为几个小节，使其更符合“长文用``/`
`划分小节”的要求。 原文拆解： 1. To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion creates an outer-accounting problem distinct from within-trajectory credit assignment: row-based outcome statistics can count one episode outcome multiple times, while token-level reduction can assign unequal aggregate objective mass across trajectories. We term these effects outcome-statistics multiplicity and objective-mass imbalance. 2. We introduce Outcome-Mass Alignment (OMA), a two-component method that makes both accounting choices explicit. Outcome Alignment computes outcome statistics over unique source trajectories, while Mass Alignment controls aggregate objective mass after within-trajectory aggregation of optimizer-native token objectives. Neither redesigns the base optimizer's credit assignment, importance ratios, or clipping. 3. Across matched ALFWorld GRPO comparisons at two model scales, where failed trajectories typically expand much more than successful ones, OMA improves matched-horizon endpoint success by 7.0-9.4 percentage points. Ordinary WebShop exhibits weak or reversed expansion-outcome coupling and correspondingly mixed effects, whereas in a success-oriented WebShop setting, WebShop-SOL, OMA improves success nAUC from 0.399 to 0.511. Component ablations and GiGPO and GSPO comparisons further support component separability and cross-optimizer compatibility. These results show that trajectory expansion always defines an implicit weighting rule, but its practical effect depends on how expansion couples to outcomes and optimizer-native updates. 划分小节： 可以加两个h2或h3。
问题背景与挑战
-> 但规则说“保留原文所有信息，不得增删改内容”。如果加小标题，算不算增改内容？ 通常加HTML结构标签不算增改文本内容，但加中文小标题会引入原文没有的文本。为了严格遵守“不得增删改内容”，小标题应该用原文中的词，或者干脆不加小标题，只用``，或者把原文的第一句或某句作为小标题？ 或者，因为原文只有一段，直接用一个`
`包裹即可，或者为了体现“长文用h2/h3划分”，我可以用原文的句子作为小标题？ 比如：
Outer-Accounting Problem
(但原文是 "outer-accounting problem"，如果直接用原文的句子或短语，可能会被认为增改。) 更安全的做法：不添加原文没有的文本。如果不需要划分，就只用``。但规则说“长文用`
`/`
`划分小节”，这可以是一个建议（如果需要的话）。原文其实是一段完整的摘要。如果非要划分，可以把原文拆成几个`
`，或者用原文的词做小标题。 比如：
Problem Definition
这属于增改内容。 不如直接全部用``，分成几个自然段。原文是一整段，我可以把它按逻辑分成几个`
`，不加小标题，因为加小标题必然引入新词。 或者，把原文的第一句话作为`
`？"To keep long interaction contexts tractable..." 这不太像标题。 其实，如果不加h2/h3，只使用p，也符合“每个自然段落用p包裹”，而“长文用h2/h3划分”是条件性的。为了绝对不增改内容，我只使用`
`和``。 让我们把原文按逻辑拆分为几个`
`：
To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion creates an outer-accounting problem distinct from within-trajectory credit assignment: row-based outcome statistics can count one episode outcome multiple times, while token-level reduction can assign unequal aggregate objective mass across trajectories. We term these effects outcome-statistics multiplicity and objective-mass imbalance.
We introduce Outcome-Mass Alignment (OMA), a two-component method that makes both accounting choices explicit. Outcome Alignment computes outcome statistics over unique source trajectories, while Mass Alignment controls aggregate objective mass after within-trajectory aggregation of optimizer-native token objectives. Neither redesigns the base optimizer's credit assignment, importance ratios, or clipping.
Across matched ALFWorld GRPO comparisons at two model scales, where failed trajectories typically expand much more than successful ones, OMA improves matched-horizon endpoint success by 7.0-9.4 percentage points. Ordinary WebShop exhibits weak or reversed expansion-outcome coupling and correspondingly mixed effects, whereas in a success-oriented WebShop setting, WebShop-SOL, OMA improves success nAUC from 0.399 to 0.511. Component ablations and GiGPO and GSPO comparisons further support component separability and cross-optimizer compatibility. These results show that trajectory expansion always defines an implicit weighting rule, but its practical effect depends on how expansion couples to outcomes and optimizer-native updates.
检查是否增删改内容： 原文： To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion creates an outer-accounting problem distinct from within-trajectory credit assignment: row-based outcome statistics can count one episode outcome multiple times, while token-level reduction can assign unequal aggregate objective mass across trajectories. We term these effects outcome-statistics multiplicity and objective-mass imbalance. 拆分后：To keep long interaction contexts tractable, stepwise group-relative training for multi-turn LLM agents materializes each source trajectory into multiple response rows and valid tokens. This expansion creates an outer-accounting problem distinct from within-trajectory credit assignment: row-based outcome statistics can count one episode outcome multiple times, while token-level reduction can assign unequal aggregate objective mass across trajectories. We term these effects outcome-statistics multiplicity and objective-mass imbalance.
内容完全一致，只是加了``和``。 第二段： We introduce Outcome-Mass Alignment (OMA), a two-component method that makes both accounting choices explicit. Outcome Alignment computes outcome statistics over unique source trajectories, while Mass Alignment controls aggregate objective mass after within-trajectory aggregation of optimizer-native token objectives. Neither redesigns the base optimizer's credit assignment, importance ratios, or clipping. 加`
`和`Outcome-Mass Alignment (OMA)`。 第三段： Across matched ALFWorld GRPO comparisons at two model scales, where failed trajectories typically expand much more than successful ones, OMA improves matched-horizon endpoint success by 7.0-9.4 percentage points. Ordinary WebShop exhibits weak or reversed expansion-outcome coupling and correspondingly mixed effects, whereas in a success-oriented WebShop setting, WebShop-SOL, OMA improves success nAUC from 0.399 to 0.511. Component ablations and GiGPO and GSPO comparisons further support component separability