double-loop-learning

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Double-Loop Learning Review

双环学习评审

Surface the hidden assumptions behind decisions, then assess whether they still hold. Single-loop learning fixes errors; double-loop questions the goals, values, and mental models that produced them.
揭示决策背后隐藏的假设,然后评估这些假设是否仍然成立。单环学习是修正错误;而双环学习则会对产生这些错误的目标、价值观和思维模式提出质疑。

Quick Reference

快速参考

StepJobOutput
1Identify input and modePre-decision (artifact) or periodic review (patterns)
2Interrogative passDraft list of falsifiable assumptions
3Contrastive deepeningSketches of alternatives -> additional assumptions missed
4Assess and build mapAssumption map + prose summary
步骤任务输出
1确定输入内容与模式预决策(工件)或定期评审(模式)
2质询梳理可证伪假设草稿清单
3对比深化替代方案草图 -> 补充遗漏的假设
4评估并构建映射表假设映射表 + 文字摘要

Process

流程

1. Identify the Input

1. 确定输入内容

Determine the mode:
  • Pre-decision — User provides an artifact (design, spec, architecture, code). Surface what it assumes.
  • Periodic review — User points to a collection of past decisions, review results, a codebase, or asks to reflect on patterns. Surface recurring assumptions.
Read everything provided. If the input is insufficient to run the interrogative pass (e.g., no artifact, codebase, or decision history), ask the user to provide or point to the specific material to review.
确定模式:
  • 预决策 — 用户提供工件(设计、规格、架构、代码)。揭示其中隐含的假设。
  • 定期评审 — 用户指出一系列过往决策、评审结果、代码库,或是要求反思模式。揭示反复出现的假设。
阅读所有提供的内容。如果输入内容不足以开展质询梳理(例如没有工件、代码库或决策历史),请要求用户提供或指向具体的评审材料。

2. Interrogative Pass

2. 质询梳理

Work through probing questions adapted to the mode.
Pre-decision (artifact-focused):
  • What is this optimizing for? What's being traded away?
  • What constraints are treated as fixed? Which are actually choices?
  • What does this assume won't change?
  • What failure modes does this implicitly accept as unlikely?
  • Who/what does this assume will behave a certain way?
Periodic review (pattern-focused):
  • What kinds of problems keep recurring? What do they have in common?
  • What approaches have been consistently chosen? What's been consistently avoided?
  • What would someone new to this project find surprising or arbitrary?
  • Which past decisions are still shaping current choices? Are they still valid?
State each assumption as a falsifiable belief (e.g., "Users will always authenticate via OAuth" not "authentication stuff"). If you can't imagine it being false, it's not specific enough. Note where in the input each assumption appears — this becomes the Source column in the map.
针对不同模式,通过针对性问题开展梳理。
预决策(聚焦工件):
  • 该方案的优化目标是什么?牺牲了什么?
  • 哪些约束被视为固定不变的?哪些实际上是可选择的?
  • 该方案假设哪些因素不会发生变化?
  • 该方案默认认为哪些故障模式发生的可能性极低?
  • 该方案假设谁/什么会以特定方式运作?
定期评审(聚焦模式):
  • 哪些问题反复出现?它们有什么共同点?
  • 一直采用的是哪些方法?一直避免的是哪些方法?
  • 刚加入项目的人会觉得哪些内容令人惊讶或不合逻辑?
  • 哪些过往决策仍在影响当前的选择?这些决策是否仍然有效?
将每个假设表述为可证伪的观点(例如:“用户始终会通过OAuth进行身份验证”,而非“身份验证相关内容”)。如果你无法想象该假设不成立的情况,说明它不够具体。记录每个假设在输入内容中的出处——这将成为映射表中的“来源”列。

3. Contrastive Deepening

3. 对比深化

For each major assumption, briefly sketch what the design or decision would look like if that assumption were false. If inverting an assumption reveals a radically different design, there's a deeper assumption the interrogative pass didn't surface. Add it to the list.
Pre-decision: Invert the assumption and sketch the alternative design. Periodic review: Invert the assumption and consider what the pattern of decisions would have been if it had been questioned earlier.
This step can be abbreviated for simple inputs but should not be skipped — it routinely finds 2-3 assumptions the questions miss.
针对每个主要假设,简要勾勒如果该假设不成立,设计或决策会是什么样子。如果反转某个假设后呈现出完全不同的设计,说明质询梳理阶段遗漏了更深层次的假设。将其添加到假设清单中。
预决策: 反转假设,勾勒替代设计方案。 定期评审: 反转假设,思考如果更早对该假设提出质疑,决策模式会有何不同。
对于简单的输入内容,此步骤可简化但不可跳过——该步骤通常能发现2-3个质询问题未覆盖到的假设。

4. Assess and Build the Map

4. 评估并构建映射表

For each surfaced assumption, assess:
  • Still Valid — Evidence supports it, no action needed
  • Questionable — Uncertain or eroding. Pair with an experiment to test it.
  • Invalidated — Evidence contradicts it. Recommend a specific change.
Present the assumption map:
| # | Assumption | Source | Assessment | Action |
|---|-----------|--------|------------|--------|
| 1 | [falsifiable belief] | [where it shows up] | Still Valid | — |
| 2 | [falsifiable belief] | [where it shows up] | Questionable | [experiment to test] |
| 3 | [falsifiable belief] | [where it shows up] | Invalidated | [recommended change] |
Follow with a prose summary: count by assessment, highest-risk assumptions, and what would change if the Questionable/Invalidated ones were addressed.
针对每个已揭示的假设,评估其状态:
  • 仍然有效 — 有证据支持,无需采取行动
  • 存疑 — 存在不确定性或合理性正在减弱。需搭配实验进行验证。
  • 已失效 — 有证据与之矛盾。建议采取具体调整措施。
呈现假设映射表:
| # | Assumption | Source | Assessment | Action |
|---|-----------|--------|------------|--------|
| 1 | [falsifiable belief] | [where it shows up] | Still Valid | — |
| 2 | [falsifiable belief] | [where it shows up] | Questionable | [experiment to test] |
| 3 | [falsifiable belief] | [where it shows up] | Invalidated | [recommended change] |
附上文字摘要:按评估结果统计数量,列出风险最高的假设,以及解决存疑/已失效假设后会带来哪些改变。

Common Mistakes

常见误区

  • Listing facts instead of assumptions. "The system uses PostgreSQL" is a fact. "PostgreSQL will handle our scale requirements" is an assumption.
  • Assumptions too vague to assess. "We assume the architecture is good" is unfalsifiable. "Current service boundaries won't need to change for 2x traffic" is assessable.
  • Skipping the contrastive step. The interrogative pass feels complete. It isn't. Inverting assumptions surfaces the ones you can't articulate — the water the fish doesn't see.
  • Marking everything Still Valid. Every non-trivial system has assumptions under pressure. If nothing is Questionable, the review wasn't rigorous enough.
  • 将事实列为假设。“系统使用PostgreSQL”是事实。“PostgreSQL能够满足我们的规模需求”才是假设。
  • 假设表述过于模糊,无法评估。“我们假设架构是合理的”是不可证伪的。“当前服务边界在流量增长2倍时无需调整”是可评估的。
  • 跳过对比步骤。质询梳理阶段看似已完成,但实际上并非如此。反转假设能揭示那些你无法清晰表述的内容——就像鱼儿看不到自己身处的水。
  • 所有假设都标记为“仍然有效”。任何非琐碎的系统都存在面临挑战的假设。如果没有“存疑”的假设,说明评审不够严谨。

Key Principles

核心原则

  • Assumptions are falsifiable beliefs. "We assume X" not "X is a consideration."
  • Questionable is the most valuable assessment. Still Valid and Invalidated are easy. Questionable assumptions need experiments, not opinions — that's where the real leverage is.
  • The contrastive step catches what questions miss. Questions surface assumptions you can articulate. Inverting surfaces ones you can't.
  • The map is the deliverable. Not the questions asked, not the contrastive sketches. Those are working artifacts.
  • 假设是可证伪的观点。表述为“我们假设X成立”,而非“X是一个需要考虑的因素”。
  • “存疑”是最有价值的评估结果。“仍然有效”和“已失效”很容易判断。存疑的假设需要的是实验,而非主观意见——这才是真正能带来突破的地方。
  • 对比步骤能发现问题未能覆盖的内容。问题能揭示你可以清晰表述的假设,而反转假设能揭示那些你无法清晰表述的内容。
  • 映射表是最终交付物。而非提出的问题或对比草图——那些只是过程工件。