# Financial Causality Review · 金融因果标注说明

## Purpose · 目的

Review what the supplied documents support. Do not estimate an objective real-world causal effect. This pilot has 35 blinded development tasks from three stories. No model answers or suggestions are shown. Your review is evidence for the research process; it does not automatically become a gold answer.

判断提供的材料支持什么，不要求估计现实世界中的客观因果效应。本轮有三个故事的 35 道开发试标题，模型答案和建议均隐藏。人工判断用于研究复核，不会自动成为最终标准答案。

## Before you start · 开始前

- Use one pseudonymous annotator ID, such as `reviewer_A`. Do not enter your real name or email.
- Read the source excerpts yourself. Do not ask an LLM to make or check your independent judgments.
- Two reviewers should first complete the first six tasks independently. Download each person's initial answers **before** discussing differences. Keep those initial files; after discussion, save any revised answers as a separate export.
- 使用一个匿名标注编号，例如 `reviewer_A`，不要填写真实姓名或邮箱。
- 独立阅读材料，不使用大模型代答或检查独立判断。
- 两位标注者先各自完成前六题。讨论前，各自下载初始答案并保留。讨论后的修改另存一份，避免覆盖最初判断。

## For each task · 每题怎么做

1. **Read both events and their dates.** The task asks whether the proposed cause affects the proposed effect, in that direction. Check each event against the documents. Mark factual support as **supported**, **unsupported**, or **unclear**. An unsupported event is not proof that no causal link exists.

   **读两个事件和日期。** 按题目给出的“原因 → 结果”方向判断。分别检查材料是否支持两个事件，选择“支持／不支持／不清楚”。某个事件缺乏支持，不等于已经证明两者无因果。

2. **Check timing and event status.** Distinguish when something happened from when it was announced. A forecast, expected loss, or revised guidance is not an actual result. Use the temporal-order field; flag a bad date or unclear timing instead of guessing.

   **检查时间和事件状态。** 区分发生时间与公告时间。预测、预计损失和调整指引不等于已发生的结果。填写时间顺序；日期错误或时间不明时，标记问题，不要猜。

3. **Choose the relationship that the evidence supports.** Use the definitions below. Sharing a date, appearing in the same paragraph, or being financially important does not by itself establish causation.

   **选择证据支持的关系类型。** 定义见下表。同一天发生、写在同一段或财务影响重大，本身都不能证明因果。

4. **Add 1–3 exact evidence quotes.** Select text in the displayed source and click **Use selection**. Keep quotations exact; do not rewrite them. If you cannot find support, say so in the reason rather than inventing a quote. Check that each selected quote belongs to the intended source.

   **添加 1–3 段原文证据。** 在页面材料中选中文字，点击 **Use selection**。保持原文，不改写。找不到支持证据时，在理由中说明，不编造引文。核对每段引文对应的来源。

5. **Write a short reason.** State the mechanism, the missing evidence, or the ambiguity in one or two sentences. For a causal relationship, mark **direct**, **mediated/indirect**, or **unclear**. For an indirect link, name the intermediate event if the material supports it. Leave unrelated fields blank when appropriate.

   **写一两句理由。** 说明作用机制、缺失证据或歧义。因果关系可选择“直接／经中间事件传递／不清楚”。间接因果只填写材料支持的中间事件。不适用的字段可以留空。

6. **Optionally record causal strength.** This is exploratory. You may leave it blank if the distinction is not defensible. **Save & next** marks this human review as complete; it does not declare the answer correct or final.

   **因果强度可选填。** 这是探索性字段，无法可靠区分时可以留空。**Save & next** 只表示本次人工复核已完成，不表示答案已被确认正确或定为最终标准。

## Relationship types · 关系类型

| Type | Plain-language meaning / 通俗解释 |
|---|---|
| `economic_causal` | A changes whether, when, or how B happens through an economic or operational mechanism. Example: a strike stops production and delays deliveries. / A 通过经济或经营机制影响 B 的发生、时间或方式。例如：罢工使生产停滞，进而延迟交付。 |
| `accounting_composition` | A is part of, or an arithmetic component of, B; the statement only establishes that accounting relationship. Example: segment revenue is included in total revenue. / A 属于 B 或是 B 的计算组成部分，材料只说明会计构成。例如：分部收入计入总收入。 |
| `disclosure_of` | One event is an announcement or report of the other. Reporting an outcome is not the mechanism that produced it. / 一个事件只是对另一个事件的公告或披露，报告结果不是产生结果的机制。 |
| `shared_cause` | The evidence identifies another event that contributes to both A and B; it does not establish A → B. / 材料指出另一个事件同时影响 A 与 B，但未证明 A 导致 B。 |
| `cooccurrence` | The material places A and B together in time or context, without supporting a mechanism or common cause. / 材料只显示两者同时或相关背景下出现，未支持作用机制或共同原因。 |
| `no_supported_link` | The supplied material does not support the proposed directed causal link. This is not a claim that no link could exist anywhere. / 所给材料不支持题目提出的定向因果关系，不代表现实中绝不可能存在关系。 |
| `uncertain` | Missing, conflicting, or ambiguous evidence prevents a defensible choice. / 证据缺失、冲突或有歧义，无法可靠选择。 |

**Optional strength codes:** A = high; B = medium; C = low; D = no causal contribution supported for the proposed direction. U means uncertain or excluded from causal-strength scoring, for example a pure accounting-composition or disclosure relationship. **U is not a fifth strength level.** Do not use D merely because an event or its evidence is unclear. These labels concern the supplied story and evidence, not stock-price impact or a measured causal-effect size.

**可选强度代码：** A = 高，B = 中，C = 低，D = 材料不支持该方向的因果贡献。U 表示不确定，或不进入因果强度评分，例如纯财务构成、纯披露关系。**U 不是第五档强度。** 不要因为事件或证据不清楚就选择 D。这些标签针对材料中的故事关系，不表示股价影响大小，也不是测量得到的因果效应量。

## Small practice examples · 简短练习示例

These are fictional examples, not answers to the pilot tasks.

以下为虚构例子，不是本轮试标题的答案。

| Material / 材料 | What to record / 怎么判断 |
|---|---|
| “A factory strike stopped production and delayed deliveries.” / “工厂罢工导致停产，并延迟交付。” | Strike → delayed deliveries: economic causation, mediated through stopped production. Cite the sentence. / 罢工 → 交付延迟：经济因果，经“停产”传递，引用该原文。 |
| “Division revenue of $2 billion is included in group revenue of $5 billion.” / “分部收入 20 亿美元计入集团收入 50 亿美元。” | Accounting composition; U, excluded from strength scoring. The sentence alone gives no causal mechanism. / 财务构成，U，不纳入强度评分；这句话本身未提供因果机制。 |
| “The company expects a loss next quarter.” / “公司预计下一季度亏损。” | It supports a forecast, not an actual loss. Flag an event that describes the loss as already realized. / 只支持预测，不支持已经亏损；如果事件写成实际亏损，应标记事件问题。 |

If you choose a numeric strength score, use the pilot's exploratory ranges: A = 80–100, B = 50–79, C = 20–49, D = 0–19. These are operational bins, not measured causal effects. Leave the score blank when you cannot defend it.

如果自愿填写分数，使用本轮探索性区间：A = 80–100，B = 50–79，C = 20–49，D = 0–19。这只是操作性分档，并非测得的因果效应；无法可靠给分时留空。

## Ambiguity is a valid result · 可以保留不确定

Flag an incorrect event, wrong date, missing source context, forecast/actual confusion, or an unclear rule. State what cannot be decided. You do not need to force a unique answer. Do not treat every accounting term as noncausal: a statement about a cost mechanism may be causal, while a statement that a cost is included in a total may only describe composition.

可以标记事件错误、日期错误、来源上下文不足、预测与实际混淆、规则不清等问题，并说明无法判断的部分。不必强行给出唯一答案。也不要把所有会计术语都视为非因果：成本产生机制可能涉及因果，而“该费用包含在某总额内”可能只是构成关系。

## Saving and returning answers · 保存与交回

- **This version saves only in your current browser's local storage. There is no server upload or central submission.** The researcher cannot see your work until you return an exported file.
- Prefer the standalone [annotation page](https://boomq-financial-crab-annotation.static.hf.space/index.html). In an embedded Hugging Face view, **Full page ↗** opens it directly.
- Download **JSON or JSONL** after the first six tasks, periodically during longer sessions, and at completion. Send the downloaded file to the researcher through the agreed channel.
- Import your backup to resume on a new device, another browser, or after storage loss. Browser autosave does not synchronize across devices or browser profiles.
- Clearing site data, private/incognito mode, storage limits, changing the site address, or a shared browser profile can lose or mix work. Use your own browser profile and keep downloaded backups. Do not share that profile with another reviewer.
- **本版本只保存在当前浏览器的本地存储中，没有服务器上传或中央提交。** 只有交回导出文件，研究者才能收到你的结果。
- 推荐使用[独立标注页面](https://boomq-financial-crab-annotation.static.hf.space/index.html)。在 Hugging Face 嵌入视图中，可点击 **独立页面 ↗ / Full page ↗**。
- 前六题完成后、长时间标注过程中及全部完成时，下载 **JSON 或 JSONL**，通过约定渠道交给研究者。
- 换设备、换浏览器或本地记录丢失后，可导入备份继续。自动保存不会跨设备或浏览器配置同步。
- 清除网站数据、无痕模式、存储空间限制、网站地址变化或共用浏览器配置，可能导致记录丢失或混在一起。请使用自己的浏览器配置并保存下载备份，不与另一位标注者共用。