# Source-Grounded Timeline From Conflicting Dated Documents

> A reusable prompt that turns dated documents with disagreeing dates into one timeline where every conflict stays visible side by side.

## Install

Copy the content below into your project:

# Source-Grounded Timeline From Conflicting Dated Documents

A reusable prompt that turns dated documents with disagreeing dates into one timeline where every conflict stays visible side by side.

## Start here

Paste this prompt into any ordinary AI chat that accepts text. Then paste your documents underneath it, one per block, using this shape:

DOC-ID: D1
TITLE:
DOCUMENT DATE (if known): UNKNOWN
TEXT:
(paste)

If a document has no DOC-ID, add the ID yourself first — the prompt will ask you to, rather than invent one. To check the output, confirm that every date row names a DOC-ID you pasted, every support phrase can be found by Ctrl+F in your own files, and no cluster silently picks a winner where sources disagree.

## What this prompt does

It extracts only statements about dated events, groups them by real-world event, and shows competing dates side by side with the exact source, plus a conflict label. Output has five parts: header counts, numbered clusters, "Documents used", "Could not place", and review questions for you. It is written for research, study and office work where minutes, contracts, logs and clips disagree.

### Inputs you need

- At least one dated document in the DOC-ID block shape above.
- A DOC-ID you assign yourself; the prompt will not create IDs, titles, authors or dates.
- Exact wording for relative or partial dates ("two weeks later", "early May") so they stay labelled imprecise.

### Permissions and limits

Nothing is sent, filed or published; the chat only assembles a draft you review by hand. The prompt cannot browse or look up dates outside your paste, does not verify completeness, and cannot add events you did not supply — it will refuse. Ambiguous dates like 03/04/2024 are kept verbatim and marked AMBIGUOUS with both readings, not converted. An illustrative example in the prompt is fictional and shows output shape only. Source reviewed; runtime not tested.

## FAQ

**Can I ask it to just decide the correct date?** No. It deliberately keeps every competing date and never chooses a winner; resolution is left to you.

**What if I paste no documents?** It replies only with the DOC-ID template and stops.

## Attribution

Original TokRepo prompt, CC BY 4.0. Reference: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>), reviewed 2026-10-04.

## Complete reusable prompt

You are helping me build a source-grounded timeline from dated documents I paste. Nothing you produce is sent, filed or shared anywhere; you only assemble a draft I will review and edit by hand.

WHO I AM AND WHAT I NEED
I have several documents (records, meeting minutes, contracts, news clips, log notes) that mention the same events with dates that do not always agree. I need one chronological timeline where conflicts are shown side by side rather than smoothed away. The output will be pasted into my own notes or report.

WHAT I WILL PASTE
Each document should arrive with a short document ID line I assign, in this exact shape:

DOC-ID: (e.g. D1)
TITLE:
DOCUMENT DATE (if known): (e.g. 2024-03-11, or UNKNOWN)
TEXT:
(paste)

If my documents have no IDs, ask me to add them before starting; do not invent IDs, dates, titles or authors.

YOUR TASK
1. Read every document and extract only statements that reference a dated event (a thing that happened, a decision, a signing, a change, a publication, a start or stop date). Ignore opinions, forecasts and undated background.
2. Build one timeline ordered by the earliest date any document gives for each event cluster. Group statements that clearly refer to the same real-world event even when their dates differ; if two statements might be the same event or might be two different events, keep them as separate clusters and mark the ambiguity.
3. For each event cluster, show every competing date with the exact document ID, and quote the smallest phrase from the source that supports that date (under 25 words, verbatim, in quotes).
4. If a document gives a date range, a season, a month-only date, or relative wording ("two weeks later"), record it exactly as written and label it imprecise; do not convert it into a precise date unless another pasted document states the conversion.
5. If two documents conflict, present both rows under the same cluster and label the conflict type: (a) different calendar dates, (b) precise vs. approximate, (c) year disagreement, (d) one source undated.
6. Never choose a winning date. If documents differ, both stay. You may add one neutral sentence per cluster describing what would be needed to resolve it (e.g. "a document dated after the repair log").

OUTPUT FORMAT
Produce, in plain Markdown:

A. A one-line header: total documents received, total event clusters, clusters with conflicts.
B. The timeline as numbered clusters. Each cluster:
   Number and short neutral label (use only words that appear in my documents; no invented names)
   Ordered list of date rows, each row: the date exactly as written | DOC-ID | short verbatim support phrase | IMPRECISE / PRECISE / CONFLICT
   Conflict box: bullet list of the conflicting rows with the conflict-type label from rule 5
   Resolution note (optional, one sentence)
C. A separate section "Documents used" listing each DOC-ID with its date status: DATED / RANGE / RELATIVE / UNDATED.
D. A section "Could not place" listing any dated statements too vague to cluster, with DOC-ID and the verbatim phrase.
E. Three to six review questions for me, phrased as checks I should perform myself, for example asking me to confirm whether DOC-IDs D2 and D5 describe the same meeting.

UNCERTAINTY AND MISSING INPUT
If a DOC-ID block is malformed or missing, list it under "Could not place" or ask me once for the missing ID rather than guessing. If a date string is ambiguous (03/04/2024), show it verbatim, mark it AMBIGUOUS, and note both readings (month/day or day/month) without choosing. If I paste no documents, reply only with the DOC-ID template and stop. If I ask you to add events not in my documents, refuse and say they are unsupported.

RULES
Quote only from my pasted text. Do not add historical context, do not browse, do not infer authorship, do not merge dates, do not round relative dates. Do not claim your timeline is complete or verified. Keep every cluster's support phrases verbatim so I can Ctrl+F them in my sources.

ILLUSTRATIVE FICTIONAL EXAMPLE (for shape only)
DOC-ID: D1 / TITLE: Repair log / DOCUMENT DATE: 2024-05-02 / TEXT: "Pump replaced 2024-04-28."
DOC-ID: D2 / TITLE: Site memo / DOCUMENT DATE: 2024-05-20 / TEXT: "Pump swap completed early May."
Expected shape: Cluster 1, label "Pump replaced": row 1 = 2024-04-28 | D1 | "Pump replaced 2024-04-28" | PRECISE; row 2 = early May | D2 | "completed early May" | IMPRECISE; conflict box = precise vs. approximate; resolution note = a document giving an exact swap date after D2 would settle the sequence.

CHECK YOUR OWN OUTPUT BEFORE RETURNING IT
- Every date row has a DOC-ID that appears in my paste.
- Every support phrase is a verbatim substring of my paste.
- No cluster contains a single chosen winner where documents disagree.
- Ambiguous dates are flagged, not converted.
- Sections B, C, D and E are present and none is empty without explanation.
If any check fails, fix it before answering.

BOUNDARIES
You are preparing a private draft timeline for my review. You are not publishing, emailing, filing, uploading or scheduling anything, and you cannot look up dates outside what I pasted. After I get your draft, I will verify each support phrase against my originals myself.

## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 基于原始出处的冲突日期时间线整理提示词

一条可复用提示词：把日期互相矛盾的带日期文档整理成一条时间线，所有冲突并列保留、不擅自取舍。

## 开始使用

把这条提示词粘贴进任何接受文本的普通 AI 聊天窗口，然后在它下面按块粘贴你的文档，每块格式如下：

DOC-ID: D1
TITLE:
DOCUMENT DATE (if known): UNKNOWN
TEXT:
(paste)

如果文档没有 DOC-ID，请先自己补上，提示词会要求你补而不是替你编造。检查输出时确认：每个日期行都对应你粘贴过的 DOC-ID，每条支持短语都能在你自己的文件里用 Ctrl+F 搜到，且没有哪个事件组在来源冲突时被悄悄选了唯一答案。

## 这条提示词做什么

它只提取带日期的“事件”陈述，按真实事件归类，并把互相竞争的日期与原文出处并列展示，同时给出冲突类型标签。输出分五部分：统计表头、编号事件组、“用到的文档”“无法归置”以及给你自查的复核问题。适用于会议纪要、合同、日志、新闻摘录日期不一致的研究、学习与办公场景。

### 需要准备的输入

- 至少一份按上述 DOC-ID 块格式提供的带日期文档。
- 由你自己指定的 DOC-ID；提示词不会生成 ID、标题、作者或日期。
- 相对或部分日期的原始写法（如“两周后”“5 月初”），以便保留“不精确”标记。

### 权限与限制

内容不会被发送、归档或发布，聊天只生成一份由你人工复核的草稿。提示词无法联网，也不能查你粘贴之外的信息，不保证时间线完整，且拒绝添加你没提供的活动。像 03/04/2024 这样的歧义日期会原样保留并标为 AMBIGUOUS，同时列出两种读法，不做换算。提示词中的示例为虚构，仅示意输出形状。来源已审阅，未做运行测试。

## 常见问题

**能不能直接让它判定正确日期？** 不能。它刻意保留所有竞争日期、绝不选唯一答案，判断留给你。

**一份文档都没粘贴会怎样？** 它只回一份 DOC-ID 模板然后停止。

## 出处说明

TokRepo 原创提示词，CC BY 4.0。参考：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)，审阅于 2026-10-04。

## 完整可复制提示词

你正在帮助我根据我粘贴的带日期文档构建一份有来源依据的时间线。你生成的任何内容都不会被发送、归档或共享到任何地方；你只是组装一份我将人工审阅并编辑的草稿。

我是谁，我需要什么
我有若干文档（记录、会议纪要、合同、新闻摘录、日志备注），它们提到相同的事件，但日期并不总是一致。我需要一条按时间顺序排列的时间线，其中冲突是并列展示的，而不是被抹平。输出将被粘贴到我自己的笔记或报告中。

我将粘贴什么
每份文档都应带有一行由我指定的简短文档 ID，格式如下：

DOC-ID: (e.g. D1)
TITLE:
DOCUMENT DATE (if known): (e.g. 2024-03-11, or UNKNOWN)
TEXT:
(paste)

如果我的文档没有 ID，请在开始前要求我添加；不要编造 ID、日期、标题或作者。

你的任务
1. 阅读每一份文档，只提取引用带日期事件的陈述（发生的事、某项决定、签署、变更、发布、开始或结束日期）。忽略意见、预测和无日期的背景信息。
2. 构建一条时间线，按任何文档为每个事件组给出的最早日期排序。将明确指涉同一真实世界事件的陈述归为一组，即使它们的日期不同；如果两条陈述可能是同一事件，也可能是两个不同事件，则将它们保留为单独的组并标记这一歧义。
3. 对每个事件组，展示每一个互相竞争的日期以及确切的文档 ID，并引用来源中支持该日期的最短短语（少于 25 词，逐字引用，加引号）。
4. 如果某份文档给出的是一个日期范围、某个季节、仅有月份的日期，或相对措辞（“两周后”），请原样记录并标记为不精确；除非另一份已粘贴的文档明确说明了换算，否则不要将其转换为精确日期。
5. 如果两份文档冲突，在同一事件组下呈现两行，并标注冲突类型：(a) 日历日期不同，(b) 精确 vs. 近似，(c) 年份不一致，(d) 一个来源无日期。
6. 绝不要选定一个获胜日期。如果文档不同，两者都保留。你可以为每个事件组添加一句中性的话，描述解决该冲突需要什么（例如“一份日期在维修日志之后的文档”）。

输出格式
以纯 Markdown 生成：

A. 一行表头：收到的文档总数、事件组总数、存在冲突的组数。
B. 时间线为编号的事件组。每个事件组：
   编号和简短中性的标签（只使用我的文档中出现的词；不要编造名称）
   按顺序排列的日期行列表，每一行：原样写出的日期 | DOC-ID | 简短逐字支持短语 | IMPRECISE / PRECISE / CONFLICT
   冲突框：以项目符号列出冲突行，并附上规则 5 中的冲突类型标签
   解决说明（可选，一句话）
C. 单独一节“用到的文档”，列出每个 DOC-ID 及其日期状态：DATED / RANGE / RELATIVE / UNDATED。
D. 一节“无法归置”，列出任何过于模糊而无法归组的带日期陈述，并附上 DOC-ID 和逐字短语。
E. 三到六个供我复核的问题，措辞为我应当自己执行的检查，例如要求我确认 DOC-ID D2 和 D5 是否描述同一场会议。

不确定性与缺少输入
如果某个 DOC-ID 块格式错误或缺失，请将其列在“无法归置”下，或向我询问一次缺失的 ID，而不要猜测。如果某个日期字符串有歧义（03/04/2024），请原样显示，标记为 AMBIGUOUS，并指出两种读法（月/日或日/月），但不要选择。如果我没有粘贴任何文档，只回复 DOC-ID 模板然后停止。如果你要求我添加我的文档中没有的事件，请拒绝并说明这些事件没有支持。

RULES
只引用我粘贴的文本。不要添加历史背景，不要联网浏览，不要推断作者身份，不要合并日期，不要对相对日期做四舍五入。不要声称你的时间线是完整或经过核实的。保留每个事件组的支持短语原文，以便我能在来源中 Ctrl+F 检索。

说明性虚构示例（仅示意形状）
DOC-ID: D1 / TITLE: Repair log / DOCUMENT DATE: 2024-05-02 / TEXT: "Pump replaced 2024-04-28."
DOC-ID: D2 / TITLE: Site memo / DOCUMENT DATE: 2024-05-20 / TEXT: "Pump swap completed early May."
预期形状：事件组 1，标签 "Pump replaced"：第 1 行 = 2024-04-28 | D1 | "Pump replaced 2024-04-28" | PRECISE；第 2 行 = early May | D2 | "completed early May" | IMPRECISE；冲突框 = 精确 vs. 近似；解决说明 = 在 D2 之后给出确切更换日期的文档才能解决先后顺序。

返回前检查你自己的输出
- 每个日期行都有一个出现在我粘贴内容中的 DOC-ID。
- 每条支持短语都是我粘贴内容的逐字子串。
- 当文档不一致时，没有哪个事件组只给出唯一选定的答案。
- 歧义日期被标记，而非换算。
- 部分 B、C、D 和 E 均存在，且没有哪个部分在无说明的情况下为空。
如果任何检查失败，在回答前修正它。

边界
你正在为我复核准备一份私密草稿时间线。你不发布、不发邮件、不归档、不上传、不安排任何事项，且你不能查询我粘贴之外的日期。在我拿到你的草稿后，我会自己核对每条支持短语与我的原件。

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/source-grounded-timeline-conflicting-dated-documents-dee661dc
Author: Prompt Lab