thinking-model-selection

模型选择

与 thinking-model-router 有重叠。 如果你已经知道哪个模型适用，就完全跳过路由，直接调用该模型。如果你还不知道，thinking-model-router 是唯一的入口——从这里开始。此技能会深入讲解 如何 对问题分类，以及选定模型后 何时放弃 它。（审计建议将此内容合并到 thinking-model-router；目前请优先将路由器作为入口。）

概述

每种心智模型都有擅长的领域，也有不适用的领域。模型选择是一项元技能，用于识别哪种模型适用于哪类问题。专家不仅了解许多模型，还知道何时应用每一种模型。用错模型往往比完全不用模型更糟。

核心原则： 地图不是疆域。选择最适合你旅程的地图。

何时使用

• 面对新的问题类型
• 当前方法不起作用
• 多个利益相关方建议使用不同框架
• 决定如何组织分析
• 教别人使用哪些工具
• 构建决策流程

决策流程：

Facing a problem?
  → Have you identified the problem type? → no → CLASSIFY THE PROBLEM FIRST
  → Does your usual model fit? → no → CONSIDER ALTERNATIVES
  → Are you using a model by habit? → yes → QUESTION THE FIT

何时不使用

• 你不知道该用哪种思维工具。 从 thinking-model-router 开始——它是将你引导至合适技能的唯一入口。此技能会深入讲解 如何 对问题分类并匹配模型，但仅在路由器缩小范围后使用。
• 你已经对问题类型完成分类。 如果你知道这是一个调试问题，直接调用 thinking-scientific-method 或 thinking-five-whys-plus 即可。不要为了确认已知结论而进行完整的模型选择流程。
• 问题很常规，且你的默认方法适用。 模型选择会增加开销；遇到熟悉的问题时，使用你知道有效的模型即可。
• 你想把它作为每项任务的第一步。 这样会让元技能变成负担。大多数任务不需要模型选择——它们需要执行。只有在你惯用的方法确实失效时，才使用此技能。

转向： 几乎所有情况下，都从 thinking-model-router 开始——它会在一次流程中对问题分类并分派到合适的技能。此技能用于说明 如何 进行分类，而不是替代路由器。

触发卡片

当你惯用的问题解决方法确实失效，需要为任务匹配其他工具时：

1. 对问题分类： 诊断 / 决策 / 理解 / 创意 / 评估。
2. 检查约束： 时间紧迫？信息缺口？风险高？复杂度高？
3. 匹配： 使用下方的问题-模型匹配表，查找类别 → 选择具体模型。
4. 设定退出标准： 如果 15 分钟以上仍无新见解，或关键事实不匹配，就切换。

如果你完全不知道该用哪种方法，请从 thinking-model-router 开始——它会在一次流程中完成分类和分派。

问题-模型匹配

步骤 1：对问题分类

## Problem Classification

Problem: [Describe the problem]

Problem dimensions:
| Dimension | Assessment |
|-----------|------------|
| Predictability | Can outcomes be predicted? [High/Medium/Low] |
| Complexity | How many interacting parts? [Simple/Complicated/Complex] |
| Time horizon | When do consequences matter? [Immediate/Short/Long] |
| Reversibility | Can decisions be undone? [Easily/With difficulty/Not at all] |
| Information | How much do you know? [Complete/Partial/Minimal] |
| Stakeholders | Who's affected? [Individual/Team/Organization/Society] |

步骤 2：匹配模型类别

## Model Category Matching

Based on classification, which category fits?

| Problem Type | Model Category | Examples |
|--------------|----------------|----------|
| Root cause unknown | Diagnostic models | 5 Whys, Scientific Method, Kepner-Tregoe |
| Decision under uncertainty | Probabilistic models | Bayesian, Expected Value, Regret Minimization |
| System behavior | Systems models | Feedback Loops, Leverage Points, Archetypes |
| Cognitive bias risk | Debiasing models | Pre-mortem, Red Team, Steel-manning |
| Resource allocation | Constraint models | Theory of Constraints, Opportunity Cost |
| Innovation/exploration | Generative models | First Principles, TRIZ, Effectuation |
| Domain classification | Meta models | Cynefin, Circle of Competence |

步骤 3：选择具体模型

## Model Selection

Category: [From Step 2]

Candidate models:
| Model | Fit Score | Strengths for This Problem | Weaknesses |
|-------|-----------|---------------------------|------------|
| [Model 1] | [1-5] | [Why it fits] | [Limitations] |
| [Model 2] | [1-5] | [Why it fits] | [Limitations] |
| [Model 3] | [1-5] | [Why it fits] | [Limitations] |

Selected model: [Choice]
Rationale: [Why this model for this problem]

模型选择矩阵

按问题类型

DIAGNOSTIC PROBLEMS (What's causing this?)
├── Known categories exist → Kepner-Tregoe (systematic analysis)
├── Need quick root cause → 5 Whys Plus (iterative drilling)
├── Hypothesis-driven → Scientific Method (test and falsify)
└── System-wide issue → Feedback Loops (find reinforcing patterns)

DECISION PROBLEMS (What should we do?)
├── High stakes, irreversible → Regret Minimization, Pre-mortem
├── Under uncertainty → Bayesian, Probabilistic Thinking
├── Resource constrained → Opportunity Cost, Theory of Constraints
├── Multiple options → Kepner-Tregoe (decision analysis)
└── Type 1 vs Type 2 → Reversibility Framework

UNDERSTANDING PROBLEMS (How does this work?)
├── Complex system → Systems Thinking, Feedback Loops
├── Human behavior → Jobs to be Done, Incentive Analysis
├── Organizational → Archetypes, Leverage Points
└── Competitive → Red Team, Game Theory

CREATIVE PROBLEMS (How might we...?)
├── Break assumptions → First Principles, TRIZ
├── Limited resources → Effectuation, Via Negativa
├── Technical contradiction → TRIZ
└── Unknown territory → Thought Experiments, Cynefin (probe)

EVALUATION PROBLEMS (Is this good?)
├── Arguments/proposals → Steel-manning, Red Team
├── Predictions → Probabilistic, Calibration
├── Longevity → Lindy Effect
├── Safety → Margin of Safety, Pre-mortem
└── Expertise fit → Circle of Competence

按领域

## Domain-Model Mapping

| Domain | Primary Models | Why |
|--------|---------------|-----|
| Debugging | Scientific Method, 5 Whys | Hypothesis-differential localization, root cause |
| Architecture | Systems Thinking, Leverage Points | Interconnections, intervention |
| Product | Jobs to be Done, Cynefin | User needs, complexity |
| Strategy | Red Team, Pre-mortem | Adversarial, risk |
| Performance | Theory of Constraints, Fermi | Bottlenecks, estimation |
| Decisions | Reversibility, Regret Minimization | Stakes assessment |
| Innovation | First Principles, TRIZ | Breakthrough thinking |
| Risk | Margin of Safety, Probabilistic | Uncertainty handling |

模型失效模式

使用错误的模型

## Model Mismatch Indicators

Signs you're using the wrong model:
- Analysis feels forced or awkward
- Key aspects don't fit the framework
- You're ignoring important factors
- Results don't match intuition consistently
- Stakeholders don't recognize the framing

Common mismatches:
| Situation | Wrong Model | Right Model |
|-----------|-------------|-------------|
| Complex adaptive system | Root cause analysis | Systems thinking |
| Simple process problem | Systems thinking | Checklist/SOP |
| Uncertain future | Detailed planning | Effectuation |
| Known domain | First principles | Best practices |
| Political problem | Technical analysis | Stakeholder mapping |

过度使用模型

## Model Overuse Patterns

"When you have a hammer, everything looks like a nail"

Signs of overuse:
- You use the same model for every problem
- You haven't learned new models recently
- Problems that don't fit are "forced" into the model
- You dismiss problems that don't fit your model

Fix: Deliberately practice with unfamiliar models
     Ask: "What model would someone else use?"
     Rotate models intentionally

模型选择流程

快速选择（< 2 分钟）

## Quick Model Selection

1. What type of problem? [Diagnostic/Decision/Understanding/Creative/Evaluation]
2. What's the constraint? [Time/Information/Stakes/Complexity]
3. What's the default model for this type?
4. Any reason NOT to use the default?
5. Proceed or reconsider

Default models by type:
- Diagnostic → 5 Whys Plus
- Decision → Reversibility check first
- Understanding → Systems Thinking
- Creative → First Principles
- Evaluation → Steel-manning

审慎选择（> 5 分钟）

## Deliberate Model Selection

1. Characterize the problem fully
2. Identify 3-5 candidate models
3. Score each on fit
4. Consider combining models
5. Select and document rationale
6. Plan to reassess if model doesn't illuminate

Selection criteria:
| Criterion | Weight | Model A | Model B | Model C |
|-----------|--------|---------|---------|---------|
| Problem fit | 30% | | | |
| Available info | 20% | | | |
| Time to apply | 15% | | | |
| Stakeholder acceptance | 15% | | | |
| My competence with model | 20% | | | |

模型选择模板

# Model Selection: [Problem]

## Problem Characterization
Type: [Diagnostic/Decision/Understanding/Creative/Evaluation]
Complexity: [Simple/Complicated/Complex/Chaotic]
Stakes: [Low/Medium/High]
Reversibility: [High/Medium/Low]
Information: [Complete/Partial/Minimal]

## Candidate Models
| Model | Fit | Strengths | Weaknesses |
|-------|-----|-----------|------------|
| | | | |

## Selected Model
Model: [Choice]
Rationale: [Why this model]

## Fallback
If selected model doesn't work: [Alternative]
Signs to switch: [Indicators]

## Application Plan
How I'll use this model:
1. [Step]
2. [Step]

元模型指南

何时使用多个模型

## Model Combination Triggers

Use multiple models when:
- Problem spans multiple types
- Single model leaves blind spots
- Stakes are very high
- Time allows deeper analysis

Combination patterns:
- Sequential: Use A to narrow, then B to decide
- Parallel: Use A and B independently, compare results
- Nested: Use A at macro level, B at micro level

何时放弃某个模型

## Model Exit Criteria

Stop using a model when:
- 15+ minutes with no insight
- Key facts don't fit the framework
- You're forcing the analysis
- Someone suggests a better fit

Don't stop just because:
- The answer is uncomfortable
- It requires more work
- Results disagree with intuition (investigate why)

验证清单

• 已对问题类型分类
• 已考虑多个候选模型
• 已检查模型与问题是否匹配
• 避免因习惯而使用模型
• 模型不起作用时有备用方案
• 能够说明该模型为何适用

关键问题

• “这究竟是哪类问题？”
• “这个领域的专家会使用什么模型？”
• “我使用这个模型是因为它适合，还是因为我熟悉它？”
• “这个模型忽略了问题的哪些方面？”
• “什么模型会给出不同的答案？”
• “我应该何时切换到其他模型？”

芒格的智慧

“你必须在头脑中建立模型。你还必须把自己的经验——包括间接经验和直接经验——排列在这张由模型构成的网格上。”

“你必须了解各大领域中的重要思想，并经常运用——所有这些思想，而不只是其中几种。”

力量不在于任何单一模型，而在于拥有多种模型，并知道何时适用每一种。只有与任务相匹配时，工具才有用。选择模型这项元技能，能倍增你所掌握的每个模型的价值。