Skip to content
Back to skills

Inductive Hypothesis Generation

ASecurity

Strategy: 从数据/观察归纳提炼假设

  • 417 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added May 27, 2026
researchgo

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned May 27, 2026

npx -y skills add yogsoth-ai/de-anthropocentric-research-engine --skill inductive-hypothesis-generation --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Inductive Hypothesis Generation?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Inductive Hypothesis Generation
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/yogsoth-ai-inductive-hypothesis-generation/badge)](https://www.skillsdirectory.com/skills/yogsoth-ai-inductive-hypothesis-generation)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: inductive-hypothesis-generation
description: "Strategy: 从数据/观察归纳提炼假设"
version: 1.0.0
category: hypothesis-formation
type: strategy
campaign: hypothesis-formulation
tactics:
  - anomaly-driven-abduction
  - falsifiability-audit
sops:
  - anomaly-characterization
  - explanation-generation
  - variable-identification
  - relationship-specification
  - falsifiability-check
dependencies:
  skills:
    - context-management
    - subagent-spawning
    - literature-engine
---

# Inductive Hypothesis Generation

从数据/观察归纳提炼假设:在理论空白或理论不足的领域,从经验模式中提炼规律,谨慎泛化为可测试命题。

## 适用场景

- 领域缺乏成熟理论,但积累了大量经验观察或数据模式
- 研究 gap 表现为"反复出现的现象尚无系统性解释"
- 目标是从数据中提炼规律,为后续理论建构奠基
- 探索性研究阶段,尚不清楚哪些变量重要

不适用:已有明确理论框架的领域 → 改用 deductive-hypothesis-generation。

## 思维框架

**Observe patterns → Extract regularity → Generalize cautiously → Formulate testable claim**

归纳的核心逻辑:

1. **Observe patterns**:系统整理已有观察、数据、案例中反复出现的模式(不是单次异常)
2. **Extract regularity**:识别模式背后的规律性——什么条件下出现,什么条件下不出现
3. **Generalize cautiously**:将规律从具体样本谨慎泛化——明确泛化的边界,不过度外推
4. **Formulate testable claim**:将泛化规律转化为可在新样本上检验的命题

**归纳的核心风险**:过度泛化(从有限样本跳到普遍规律)。每个归纳假设必须明确:
- 观察来自哪些样本(样本特征、来源、时间范围)
- 泛化到哪些群体(泛化范围的边界)
- 什么证据会限制或推翻泛化

## Budget Gate

| Tier | 模式覆盖 | 规律提取 | 假设产出 | 泛化边界 |
|------|---------|---------|---------|---------|
| S | ≥3 个独立观察模式 | ≥2 条规律 | ≥2 个结构化假设 | 每个假设明确样本来源 |
| M | ≥5 个独立观察模式 | ≥3 条规律 | ≥3 个结构化假设 | 泛化边界 + falsification scenario |
| L | ≥8 个独立观察模式 | ≥5 条规律 | ≥4 个结构化假设 | 完整泛化边界 + 竞争规律比较 |

## 默认参考流

1. 调用 `anomaly-characterization` SOP:系统整理已有观察/数据中的模式(含频率、条件、例外)
2. 调用 `explanation-generation` SOP(via `anomaly-driven-abduction` tactic):为每个模式生成候选规律解释
3. 调用 `variable-identification` SOP:将规律中的构念转化为可操作变量
4. 调用 `relationship-specification` SOP:明确变量间方向性关系(含调节条件)
5. 调用 `falsifiability-check` SOP(via `falsifiability-audit` tactic):为每个假设生成 falsification scenario + 泛化边界

## context-checkpoint

每轮结束后记录:
- 已整理的观察模式清单(模式描述、来源、出现频率)
- 提取的规律列表(规律陈述、支撑模式、例外情况)
- 当前假设草稿集(含泛化边界声明)
- 可证伪性状态 + 过度泛化风险评估

Files in this skill

  • SKILL.md3.2 KB
  • prompt.md4.4 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.