5 Whys Analysis — First Principles Thinking Framework

A systematic method to break down any problem/decision/belief to its fundamental truths, then rebuild from those truths to derive actionable steps.

Core Philosophy

Most people reason by analogy — "others do this, so I should too" or "this worked before, so it will work again."
This framework reasons by first principles — decomposing to irreducible facts, then reconstructing from there.

The goal is not to find "the right answer" but to surface hidden assumptions and understand the structure beneath a situation well enough to act with clarity.

The Iron Law

NO CONCLUSIONS WITHOUT REACHING IRREDUCIBLE TRUTHS

If you haven't completed the downward movement (Steps 1-4), you cannot propose conclusions or actions. Skipping to solutions without reaching fundamental truths is guessing, not analysis.

This rule applies especially when:

• The answer seems obvious (obvious answers have hidden assumptions too)
• Under time pressure (urgency makes skipping tempting)
• The user asks "so what should I do?" before analysis is done (hold your ground)

Analysis Depth Grading

Different problems need different depths. Use this guide to determine how many Whys to go:

| Level | Depth | Suitable for | Typical Why count | Stop when | |---|---|---|---|---| | L1: 表层分析 | 现象→直接原因 | 一天内的操作问题、简单技术故障 | 1-2 Why | 找到可操作的原因即可 | | L2: 归因分析 | 现象→假设→原因 | 重复出现的问题、中等复杂的决定 | 3-4 Why | 找到可验证的因果链 | | L3: 本质分析 | 现象→假设→机制→底层事实 | 长期策略、重大决策、反复出现的困境 | 5+ Why | 到达不可分事实 | | L4: 第一性原理 | 完整拆解到最底层再重构 | 颠覆性创新、突破性选择、核心信念质疑 | 直到无法再拆 | 物理/数学/逻辑极限 |

Default: start at L2. If the analysis feels shallow or the answer still doesn't hold, go deeper.

Forbidden Responses

NEVER say these in an analysis:

| 禁止的表达 | 为什么错 | |---|---| | "这个问题的本质是..."（直接跳到 Step 4，跳过了假设和机制） | 跳过假设直接给结论，很可能只是你的直觉不是本质 | | "XX 理论/模型告诉我们..."（用权威术语代替思考） | 套用框架不是分析，只是换了种说法描述现象 | | "所以你应该做 X"（在完成向下重建之前给建议） | 没有底层事实支撑的建议和碰运气没有区别 | | "这很简单，就是..."（预判问题简单会让人停止深挖） | 简单可能是熟悉，不一定是理解 | | "大多数人都是这么想的"（用共识代替论证） | 共识不等于正确，从众效应本身就是需要拆解的假设 | | "根据历史数据/经验来看..."（以外推代替分析） | 过去成立不代表未来成立，外推本身是需要验证的假设 |

Anti-Patterns

| 反模式 | 表现 | 纠正 | |---|---|---| | 锚定式分析 | 心里先有了答案，反推"为什么"来验证它 | 先列假设再问为什么，而不是先有结论再找理由 | | 挖歪了还继续挖 | Whys 已经偏离原问题，但不停下来纠偏 | 每步都回头看一眼：这个 Why 和原始问题还有关系吗？ | | 混淆因果关系和相关性 | A 和 B 同时发生 → A 导致 B | 每层 Why 后都反问：这个因果方向确定吗？有没有反向的可能？ | | 过早收敛 | 挖到第一个看起来合理的答案就停 | 列至少 2-3 条并行的 Why 链，选最深的那个 | | 用机制代替底层 | 挖到"因为市场机制如此"就停 | 继续问：这个机制为什么存在？它的前提是什么？ | | 重建时忘记约束条件 | 向上推导时添加了底层事实没有支撑的新假设 | 每一步向上推导只使用上一步已有的信息，不引入新假设 |

Evidence Requirement

Every step must produce verifiable evidence before proceeding to the next:

| Step | Evidence Required | Format | |---|---|---| | Step 1: 现象 | 原始表述引用或观察记录 | 引用原话，或"观察到 X 发生了 N 次" | | Step 2: 假设 | 每条假设标注依据 | "因为 A 曾经成立，所以假设它对 B 也成立（类比推理）" | | Step 3: Why 链 | 每个 Why 必须有事实/数据/推理支撑 | "Why？因为数据显示 X → 证据：3 月份的报表中 Y=5.2%" | | Step 4: 底层事实 | 标注该事实的不可分类型 | 物理定律/数学约束/定义真理/基本需求 | | Step 5: 约束条件 | 每条约束直接引用底层事实编号 | "约束 1：来自事实 1 → 不可能同时满足 A 和 B" | | Step 6: 可行方向 | 每个方向标注"基于哪些约束" | "方向 2 基于约束 1 和约束 3" | | Step 7: 最小动作 | 描述可观测的成功/失败标准 | "如果 X 发生后 Y > 5，则假设成立；否则需要重新审视 Step 3" |

If evidence is missing at any step, stop and flag it: "缺少证据: 无法确认 Why 3 是否成立，需要先查 X 数据."

Completing the Analysis: Next Steps

After finishing the 5 Why analysis, automatically transition to the appropriate next skill:

| Analysis Result | Next Skill | What it does | |---|---|---| | 分析指向一个可实施的方案 | writing-plans | 将方案拆成可执行的任务列表 | | 分析发现需要先验证某个前提 | task（新对话） | 创建独立验证任务，验证结果再继续 | | 分析发现是技术问题/故障 | systematic-debugging | 进入技术根因排查流程 | | 分析结果需要存档或分享 | 直接写文档 | 将分析结果写成结构化文档保存 |

Transition rule: Don't invoke the next skill until the user has confirmed they're satisfied with the analysis. Present the analysis result, then say: "根据分析结果，建议下一步用 [技能名] 推进，需要我开始吗？"

The Framework: Two Movements

                    DOWNWARD MOVEMENT
     ┌─────────────────────────────────────────┐
     │  1. 现象 (Phenomenon)                    │
     │     The surface-level observation        │
     │         ↓ "Why?"                         │
     │  2. 默认假设 (Default Assumptions)        │
     │     What you implicitly believe           │
     │         ↓ "Why?"                         │
     │  3. 机制/结构 (Mechanism / Structure)     │
     │     How it actually works                 │
     │         ↓ "Why?"                         │
     │  4. 底层事实 (Irreducible Truth)          │
     │     Can't be broken further               │
     └─────────────────────────────────────────┘

                    UPWARD MOVEMENT
     ┌─────────────────────────────────────────┐
     │  4. 底层原理 (Fundamental Truth)          │
     │     The base reality                      │
     │         ↑ "Therefore..."                  │
     │  5. 约束条件 (Constraints)                │
     │     What's possible/impossible            │
     │         ↑ "Therefore..."                  │
     │  6. 可行方向 (Feasible Directions)        │
     │     Options consistent with reality       │
     │         ↑ "Therefore..."                  │
     │  7. 最小动作 (First Minimal Action)       │
     │     What to do next                       │
     └─────────────────────────────────────────┘

Step-by-Step Method

Downward Movement: Deconstruct

Step 1: Identify the surface claim or observation

State the exact thing the user is questioning or evaluating. Write it down as-is.

Step 2: Surface default assumptions

Ask: "What does this person/statement implicitly believe to be true?"

Common patterns of default assumptions:

• 类比推理 (Analogy): "X worked for Y, so it should work for me"
• 历史外推 (Historical extrapolation): "The trend has held for N years, so it will continue"
• 权威依赖 (Authority dependence): "Experts say this, so it must be true"
• 从众效应 (Bandwagon): "Everyone else is doing this, so it's the right choice"
• 幸存者偏差 (Survivorship bias): "I only see successful examples"
• 单一归因 (Single cause): "This one factor explains everything"

List all hidden assumptions explicitly. Label each by type.

Step 3: Apply 5 Whys — iterate until irreducible

For each assumption, ask "why is this true?" then ask "why" again on the answer. Continue until you reach a statement that:

• Is a physical law (e.g., "you can't be in two places at once")
• Is a basic human need (e.g., "people need safety and belonging")
• Is a definitional truth (e.g., "if returns are uncertain, it's not insurance")
• Is a mathematical/logical constraint (e.g., "a portfolio can't have both guaranteed returns and unlimited upside")
• Is a fact that no reasonable person would dispute (e.g., "you have 24 hours per day")

When to stop:

• The answer is self-evident and needs no further support
• Further "why" would lead to speculation with no evidence
• You've reached a principle that applies across many domains (indicating you've found a universal truth)

Step 4: Record irreducible truths

Write down each irreducible truth as a standalone statement. These are your building blocks. Everything above is derived from these.

Upward Movement: Reconstruct

Step 5: Identify constraints

From each irreducible truth, derive what is necessarily true as a consequence:

• What actions are ruled out by this truth?
• What actions are required by this truth?
• What trade-offs are inherent (can't have both A and B)?

Step 6: Enumerate feasible directions

Based on constraints, list all possible approaches that don't violate the fundamental truths. At this stage, don't judge — just generate options consistent with reality.

Step 7: Design the first minimal action

For each feasible direction, define the smallest possible test that would:

1. Validate whether your understanding of the truth is correct
2. Generate real-world feedback with minimal cost/time

A good minimal action:

• Takes less than 1 hour
• Costs little to nothing
• Produces observable, unambiguous data
• Can be done today or tomorrow

Present the action concretely: "Do X, then observe Y, then decide Z."

Complementary Lenses

These are additional frameworks that can be layered on top of the 5 Whys when you want even deeper analysis. Use them if the 5 Whys alone feels insufficient for the specific problem.

System Thinking

Ask: "What system is this part of?"

• Identify the elements (people, things, resources)
• Identify the connections (how elements interact)
• Identify feedback loops (what amplifies or dampens behavior)
• Find leverage points — where a small change produces a large effect

Second-Order Thinking

Think beyond the immediate effect:

1. First order: "If I do X, what happens?"
2. Second order: "If everyone else also does X (because they're thinking the same way), what happens?"
3. Check for crowding effects — popular strategies tend to self-destruct

Anti-Fragile Thinking

Ask: "Does this benefit or suffer from volatility/uncertainty?"

| Classification | Behavior under randomness | Example | |---|---|---| | Fragile | Breaks under shock | Single income source, one-platform business | | Robust | Unaffected | Cash, diversified holdings | | Anti-fragile | Gets stronger | Options, community, skills that compound |

Assumption Flipping

Take each assumption and ask: "What if the opposite were true?"

• This reveals the boundary conditions of your reasoning
• Often surfaces a third option you hadn't considered

Output Format

When applying this method, structure the response as follows. Every item must include evidence annotation in brackets.

## Step 1: Surface Claim
[Restate the exact thing being analyzed, cited verbatim if possible]

## Step 2: Default Assumptions
- [Assumption 1] (type: analogy) [evidence: based on X's success story]
- [Assumption 2] (type: authority) [evidence: expert Y's statement on Z]

## Step 3: 5 Whys → Irreducible Truths
[Why chain 1 with evidence at each step]
  Why? [evidence] → Why? [evidence] → ... → **Irreducible truth** (type: physical law/math/definition/need)
[Why chain 2]...

Depth level applied: L[X]

## Step 4: Constraints (from irreducible truths)
- [Constraint 1] ← from Truth 1
- [Constraint 2] ← from Truth 1 + 2

## Step 5: Feasible Directions
- [Direction 1] (based on: constraints 1, 3)
- [Direction 2] (based on: constraints 2)

## Step 6: First Minimal Action
[Action description]
Evidence to observe: [what constitutes success/failure]
Time cost: [estimated]

## Next Step (after user confirmation)
[Based on analysis result, suggest: writing-plans / systematic-debugging / document / verification]

## Bonus: Complementary Lens (optional)
[If applicable, add one lens that sharpens the analysis]

When to NOT use this

• When the user just wants empathy, not analysis (listen first)
• When the problem is purely emotional and doesn't benefit from deconstruction
• When the user explicitly asks for a quick answer, not a deep dive
• For extremely simple decisions (what to eat for lunch) — overkill