任务：审阅长文与版本变化，保留原文位置Playbook: review document changes with evidence

这组短材料是长文审阅的缩小练习，CC BY 4.0；不是公司真实制度，不是法律建议，也不代表已审阅真实合同。先使用两版原文，再看参考问题表。

• policy-v1.md
• policy-v2.md
• reference-findings.csv
• reference-summary.md

把方法扩展到长文时仍需保留完整版本、目录和跨段引用；不能凭这个短例子认定某工具能处理任意长度文件。

This is a small-scale exercise for document review under CC BY 4.0, not a real company policy, legal advice or a reviewed real contract. Start with the two versions before opening references.

• policy-v1.md
• policy-v2.md
• reference-findings.csv
• reference-summary.md

For long documents, retain complete versions, outlines and cross-references. This short example establishes no tool’s maximum usable document length.

只问四件事：哪些条款变了；数字和单位是否一致；条件/否定是否改变责任；新版内部是否冲突。交付问题矩阵：finding_id、两版位置、原句/释义、变化类别、影响范围、待确认问题、处理状态。摘要中的每条结论要对应矩阵 ID。

定位使用版本名 + P1–P6 小节号，必要时补短引文和上下文。W3C 的文本定位模型说明了用文本与上下文定位片段的方式；本练习采用易读的版本和段落号，不宣称实现了该协议。

Ask four questions: which clauses changed; whether numbers/units agree; whether conditions/negation change responsibilities; and whether the draft contradicts itself. Deliver a matrix with finding ID, both version locations, passage/paraphrase, change type, scope, question and disposition. Link each summary claim to an ID.

Use version plus section IDs P1–P6, adding a short quotation/context when useful. W3C’s text-selector model describes locating passages through text/context. This exercise uses readable version/section references and does not claim protocol implementation.

1. 固定文件版本和审阅日期。列出 P1–P6，对齐同主题条款；不要只按页码配对。
2. 先做变更表。P1/P4 未变，也要保留“未变”结论，避免只展示戏剧性变化。
3. 对数字保留单位：14 calendar days 与 10 business days 不能直接按 14 和 10 比大小。缺节假日日历时暂不换算。
4. 单独读 P3 的否定与例外，把条件拆成到期前、到期后两种情况；不替作者选择解释。
5. 汇总所有涉及同一主题的段落。新版 P5 与 P6 的保存天数不一致，不能因为 P6 是摘要就忽略冲突。
6. 输出待确认问题和可追溯摘要。未获确认的问题保留开放，改写稿另标草稿。

1. Fix file versions and review date. Align sections by topic, not page number alone.
2. Record changes and unchanged clauses. P1/P4 remain unchanged.
3. Preserve units: 14 calendar days and 10 business days cannot be compared as bare 14 and 10. Do not convert without a holiday/calendar rule.
4. Read P3’s negation/exception separately, considering before and after the due date. Do not choose an interpretation for the author.
5. Cross-check every section on the same topic. New P5/P6 disagree on retention days; a summary does not cancel the conflict.
6. Return open questions and a traceable summary. Keep unresolved findings open and label any rewrite as a draft.

审阅两版虚构设备借用规则，不提供法律结论。
按小节对齐，列出变化、未变内容和新版内部冲突。
每条发现给版本、小节、支持原句或准确释义。
保留数字单位、否定和例外；不补作者意图。
先输出问题矩阵，再输出带 finding_id 的摘要。
不修改原文件，不把未确认条款标成已批准。

真实长文可先建立全篇索引，再分段审阅，最后专门查跨段关系；分段后的摘要不能替代原文。

Review the two fictional equipment-loan policies without legal conclusions.
Align sections; record changes, unchanged content and draft inconsistencies.
Give version, section and a supporting passage or faithful paraphrase for each finding.
Preserve units, negation and exceptions; do not invent author intent.
Return a findings matrix first, then a summary linked by finding_id.
Do not alter source files or mark unresolved clauses as approved.

For real long documents, index the whole text, review sections and perform a separate cross-section check. Section summaries never replace originals.

• F01：指出 P2 单位变化，保留 P6“two weeks”的歧义；不武断判断期限缩短了。
• F02：定位 P3 的审批条件变化，提出谁在何时审批的问题；不把双重否定丢掉。
• F03：新版 P5 为 60 days，P6 为 30 days，记录内部冲突，不自行选一个。
• P1 范围与 P4 免费未变；摘要没有虚构收费变化。
• 每条发现能打开对应版本并定位；摘要引用正确 finding_id，未确认事项仍未确认。

参考表是预先设计的审阅要点，不是产品得分表。可以拆分问题，但证据、范围和处理状态必须保留。

• F01: identify P2’s unit change and ambiguity in P6’s “two weeks”; do not assert a shorter duration.
• F02: locate P3’s approval-condition change and ask who approves when; retain the negation.
• F03: new P5 says 60 days, P6 says 30; record the conflict without selecting one.
• P1 scope and P4 free-of-charge status remain unchanged; invent no price change.
• Every finding locates its original, and summary finding IDs agree. Open issues stay open.

These designed reference points are not a product scorecard. Findings may be split if evidence, scope and disposition remain intact.

• 只得到泛泛摘要：要求逐条 finding_id 与出处，先缩到一对小节。
• 引文找不到：检查版本、遗漏的附件、OCR 或检索分段；看 引用排查。
• 单段都对、合起来冲突：建立同主题索引后再跨段查；看 检索排查。
• 被正文中的指令带走：把文档当待审数据，权限限制为读取，不执行其内嵌命令。

• Generic summary: request finding IDs/evidence, starting with one section pair.
• Missing quotation: check versions, attachments, OCR and retrieval chunks; see citation troubleshooting.
• Locally correct but globally conflicting: index related topics, then check across sections; see retrieval troubleshooting.
• Embedded document instructions redirect the task: treat the document as data and limit permissions to reading.

把确认后的发现整理成 带来源的研究简报；缺证据的结论先保留在问题表。

2026-11-04 核对了 W3C 文本定位说明。示例原文与参考审阅由本站编写；未宣称完成真实合同审查或运行任何长文产品。

Turn confirmed findings into a sourced research brief; retain unsupported conclusions as questions.

W3C text-location guidance was checked on 2026-11-04. The original sample and reference review were written for this site; no real contract review or document-product run is claimed.