`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。
正文内容：
"Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer."
分析正文，这是一段学术新闻/摘要，没有明显的列表、引用、代码。可以划分小节或者直接作为段落。
为了使用``或`
`，我可以将内容分为几个小节，比如：
小节1：背景与问题（One-Sided Verifiability）
小节2：方法与优化（Negative-Only Policy Optimization）
小节3：实验与结果（实验与结果）
但是规则要求“不得增删改内容”，如果加了小标题，算不算“增改内容”？
“保留原文所有信息，不得增删改内容”通常指正文的文本内容不能改。如果添加小标题，可能会被视为增加了内容。为了严格遵守“不得增删改内容”，我最好不添加原文没有的小标题，或者把原文的某句话作为小标题？不，原文没有标题，直接全部用`
`包裹最安全，或者将原文按逻辑分成几个`
`。
原文是一整段，我可以把它分成几个`
`，或者就作为一个`
`。
为了体现排版，我可以把重要的概念用``加粗。
比如：
Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks.
To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses.
Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer.
检查规则：
1. 只输出HTML片段本身，不要Markdown围栏、不要解释； -> 满足
2. 每个自然段落用包裹，长文用
/
划分小节； -> 这里没有小节，所以只用
，或者为了用
`，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。 正文内容： "Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer." 分析正文，这是一段学术新闻/摘要，没有明显的列表、引用、代码。可以划分小节或者直接作为段落。 为了使用``或`
`，我可以将内容分为几个小节，比如： 小节1：背景与问题（One-Sided Verifiability） 小节2：方法与优化（Negative-Only Policy Optimization） 小节3：实验与结果（实验与结果） 但是规则要求“不得增删改内容”，如果加了小标题，算不算“增改内容”？ “保留原文所有信息，不得增删改内容”通常指正文的文本内容不能改。如果添加小标题，可能会被视为增加了内容。为了严格遵守“不得增删改内容”，我最好不添加原文没有的小标题，或者把原文的某句话作为小标题？不，原文没有标题，直接全部用`
`包裹最安全，或者将原文按逻辑分成几个`
`。 原文是一整段，我可以把它分成几个`
`，或者就作为一个`
`。 为了体现排版，我可以把重要的概念用``加粗。 比如：
Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks.
To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses.
Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer.
检查规则： 1. 只输出HTML片段本身，不要Markdown围栏、不要解释； -> 满足 2. 每个自然段落用包裹，长文用
/
划分小节； -> 这里没有小节，所以只用
，或者为了用
/
，能否把原文的开头作为标题？不，不要增加内容。直接分段即可。
3. 列表内容用
- ，引用用
，代码用
，重要句子可用； -> 使用了
4. 保留原文所有信息，不得增删改内容； -> 满足，没有增删字词，只是加了HTML标签和分段。
5. 不要输出//等文档级标签。 -> 满足
再审视一下分段，原文是一整个自然段，如果强行拆成三个，算不算修改内容？“每个自然段落用
包裹”，原文只有一个自然段落，所以整个用一个
包裹是最严格符合“不增删改内容”和“每个自然段落用
包裹”的。
但是，如果只用一个
，排版效果可能不够丰富。不过规则说“长文用
/
划分小节”，这篇不算特别长，且没有现成的小标题。为了安全，我就用一个
，里面用加粗重要概念。或者分成几个
，因为段落划分属于排版，不属于内容修改。学术摘要通常是一整段，我把它分成三个逻辑段也是可以的，但最稳妥的是保持原样，就一个
。 等一下，如果我把它分成几个
，算不算“改内容”？通常排版允许换行分段。但我还是保持原段落结构，用一个`
`包裹整段，中间用``突出重点，这样绝对不涉及内容增删改。 让我们再读一遍原文，确认没有遗漏字符。 "Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer." 加粗部分： Reinforcement learning with verifiable rewards (RLVR) one-sided verifiability (OSV) Negative-Only Policy Optimization (NOPO) pass@k 构建HTML：
Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. A
，代码用，重要句子可用； -> 使用了 4. 保留原文所有信息，不得增删改内容； -> 满足，没有增删字词，只是加了HTML标签和分段。 5. 不要输出//等文档级标签。 -> 满足 再审视一下分段，原文是一整个自然段，如果强行拆成三个，算不算修改内容？“每个自然段落用
包裹”，原文只有一个自然段落，所以整个用一个
包裹是最严格符合“不增删改内容”和“每个自然段落用
包裹”的。 但是，如果只用一个
，排版效果可能不够丰富。不过规则说“长文用
/
划分小节”，这篇不算特别长，且没有现成的小标题。为了安全，我就用一个
，里面用加粗重要概念。或者分成几个
，因为段落划分属于排版，不属于内容修改。学术摘要通常是一整段，我把它分成三个逻辑段也是可以的，但最稳妥的是保持原样，就一个
。 等一下，如果我把它分成几个
，算不算“改内容”？通常排版允许换行分段。但我还是保持原段落结构，用一个`
`包裹整段，中间用``突出重点，这样绝对不涉及内容增删改。 让我们再读一遍原文，确认没有遗漏字符。 "Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. Across unit-test, constraint-check, and LLM self-verifiers, NOPO improves pass@k over baselines on code, logic, and math benchmarks when oracle rewards are unavailable. These gains persist under inference-time scaling and cross-benchmark transfer." 加粗部分： Reinforcement learning with verifiable rewards (RLVR) one-sided verifiability (OSV) Negative-Only Policy Optimization (NOPO) pass@k 构建HTML：
Reinforcement learning with verifiable rewards (RLVR) typically relies on a verifier that can reliably judge whether sampled solutions are correct. However, in many realistic settings, complete verification is generally unavailable, while cheap checks can only falsify failures. We formalize this regime as one-sided verifiability (OSV): verified negatives are reliable counterexamples, but verified positives are merely non-falsified candidates and may contain many false positives. This asymmetry makes standard RLVR unreliable, because reinforcing verified positives can directly reward shortcuts that pass partial checks. To address this issue, we propose Negative-Only Policy Optimization (NOPO), which treats verified positives as unlabeled and updates the policy only by suppressing trusted verified negatives. Specifically, NOPO combines a reference-relative negative suppression loss with a group-level adapter that scales each update by the proportion of non-falsified responses. A