abductive-hypothesis-generation
$
npx mdskill add yogsoth-ai/de-anthropocentric-research-engine/abductive-hypothesis-generation面对异常的最佳解释推理:当观察到现有理论无法解释的异常现象时,系统生成候选解释并选出最合理者作为假设。
SKILL.md
.github/skills/abductive-hypothesis-generationView on GitHub ↗
---
name: abductive-hypothesis-generation
description: "Strategy: 面对异常的最佳解释推理"
version: 1.0.0
category: hypothesis-formation
type: strategy
campaign: hypothesis-formulation
tactics:
- anomaly-driven-abduction
sops:
- anomaly-characterization
- explanation-generation
- plausibility-ranking
- falsifiability-check
dependencies:
skills:
- context-management
- subagent-spawning
- literature-engine
---
# Abductive Hypothesis Generation
面对异常的最佳解释推理:当观察到现有理论无法解释的异常现象时,系统生成候选解释并选出最合理者作为假设。
## 适用场景
- 观察到明确的异常现象(与现有理论预测不符的结果)
- 现有理论无法充分解释某个已知现象
- 需要在多个竞争解释中选出最值得检验的一个
- 研究起点是"这个结果很奇怪,为什么?"
不适用:没有明确异常、只是想探索一个新领域 → 改用 inductive-hypothesis-generation。
## 思维框架
**Anomaly → Generate candidate explanations → Rank by plausibility → Best explanation = hypothesis**
溯因推理的核心逻辑:
1. **Anomaly**:精确描述异常——什么现象、与什么预期不符、偏差有多大
2. **Generate candidate explanations**:系统生成所有能解释该异常的候选解释(不过早筛选)
3. **Rank by plausibility**:按可信度排序——哪个解释最简洁、最与已知事实一致、最可检验
4. **Best explanation = hypothesis**:选出最合理的解释作为工作假设,其余作为竞争假设保留
**溯因的核心原则**:
- **奥卡姆剃刀**:在解释力相当时,优先选择假设更少的解释
- **一致性**:最佳解释不应与其他已知事实矛盾
- **可检验性**:最佳解释必须能产生可观测的预测(否则无法验证)
- **生成完整性**:在排序前必须穷举候选解释,避免过早收敛
## Budget Gate
| Tier | 异常描述 | 候选解释 | 假设产出 | 竞争假设 |
|------|---------|---------|---------|---------|
| S | 1 个精确描述的异常 | ≥2 个候选解释 | 1 个最佳解释假设 | ≥1 个竞争假设保留 |
| M | 1–2 个异常 | ≥3 个候选解释 | ≥2 个结构化假设 | 完整可信度排序 |
| L | ≥2 个相关异常 | ≥5 个候选解释 | ≥3 个结构化假设 | 完整排序 + 区分性预测设计 |
## 默认参考流
1. 调用 `anomaly-characterization` SOP:精确描述异常(现象、预期、偏差、已排除的平凡解释)
2. 调用 `explanation-generation` SOP(via `anomaly-driven-abduction` tactic):系统生成候选解释(不过早筛选)
3. 调用 `plausibility-ranking` SOP:按简洁性、一致性、可检验性对候选解释排序
4. 调用 `falsifiability-check` SOP:为最佳解释生成 falsification scenario,确认其可检验性
## context-checkpoint
每轮结束后记录:
- 异常描述(精确版本,含偏差量化)
- 候选解释清单(含已排除的平凡解释及排除理由)
- 可信度排序结果(含排序依据)
- 最佳解释假设 + 竞争假设列表
- 区分性预测(什么实验能区分最佳解释与竞争解释)
More from yogsoth-ai/de-anthropocentric-research-engine
- ablation-brainstormRemove components one by one, observe system changes to reveal hidden dependencies and generate ideas from structural gaps.
- ablation-component-mappingMap system architecture to ablatable units for ablation studies
- ablation-designDesign ablation studies to isolate component contributions in ML systems
- ablation-executionRemove components one by one from a system, record the response/impact of each removal.
- abp-vulnerability-classificationClassify assumptions on 2 axes — load-bearing (how much conclusion depends on it) × vulnerable (how likely to be false). Focuses attention on High-Load × High-Vulnerable quadrant.
- abstraction-extractionExtract abstract principles from concrete domain cases. Strips domain-specific details to reveal transferable mechanisms.
- abstraction-ladderPerform bisociation at multiple abstraction levels
- abstraction-ladderingMove between concrete and abstract framings — 3 levels up (Why?) and 3 levels down (How?) to find the most productive research level.
- abstraction-to-designAbstract biological principle to design principle. Bridge from biology to engineering.
- action-priority-matrixCompute Risk Priority Number (RPN = S x O x D), classify failure modes into H/M/L action priority per AIAG-VDA tables.