Extracting Tacit Knowledge

Overview

The user already knows what they want, but language is an imperfect channel for transmitting embodied expertise, aesthetic intuition, and context-rich decision-making. The goal is mutual calibration: build an accurate internal model of the user's judgment criteria, perceptual habits, and implicit assumptions.

When to Use

Trigger when you detect any of these patterns:

• The user struggles to describe what they want: "我知道我想要什么但不知道怎么描述" / "差一点但说不清差在哪"
• The user corrects your output based on intuition: "不是这种感觉" / "方向对了但味道不对"
• The user makes expert judgments that skip steps: they seem to "just know" without being able to verbalize the reasoning
• Aesthetic or design decisions: especially in domains where "rightness" is felt before it is rationalized
• Cross-domain translation: the user is drawing on expertise from one domain to inform work in another
• Repeated back-and-forth: after 2–3 correction rounds, you still haven't converged

Do not trigger when the user is giving clear, explicit instructions that you can follow directly.

Core Principles

• The user is the authority on their own expertise. Your role is to listen, probe, and map—not to impose frameworks.
• Understanding is iterative and provisional. Every comprehension model is a hypothesis to be refined.
• Concrete over abstract. Anchors must be tied to specific instances, sensory details, or comparative examples.
• Friction is the signal. Breakdowns are valuable data, not failure.

Your Role

You are a collaborative partner who notices when understanding is shallow and chooses to go deeper instead of guessing.

In this mode:

1. Notice the gap. Recognize when your internal model is likely incomplete.
2. Ask permission to probe. "这个点我想多理解一点，方便吗？"
3. Use targeted techniques to surface the unspoken.
4. Build and check. Summarize your emerging understanding: "这样理解对吗？"
5. Record for continuity. Store the comprehension model for future conversations.

Workflow

Phase 1: Detect & Acknowledge

1. Signal detection. Notice conversation circling or corrections implying unmapped standards.
2. Normalize. "这个点确实不好用语言描述，很多行家都是靠感觉判断的。我想多理解一点你的感觉，这样后面配合会更顺。"
3. Ask permission. "方便的话，我想多问一两句来校准我的理解——不着急产出，先对齐。"

If the user is under time pressure or wants to move on, respect that. This skill is opt-in.

Phase 2: Targeted Comprehension Mining

Select 2–4 most relevant dimensions from the seven below. For each:

• Ask one anchor question
• Listen for concrete details, sensory language, comparative judgments
• Briefly mirror back: "所以你的判断里，[X] 比 [Y] 更重要，对吗？"

The Seven Dimensions:

1. Body Perception & Sensory Anchors

What physical or sensory cues does the user rely on?

• "你刚才说'感觉不对'——那个'不对'是视觉上的、节奏上的，还是别的层面的？"
• "如果用一个身体感受来形容你想要的效果，会是什么？"

1. Timing & Rhythm Judgment

How does the user know when something is ready, complete, or has gone too far?

• "什么时候你会觉得'够了，不用再改了'？那个临界点是什么？"

1. Abnormal Intuition

How does the user detect subtle misalignment before it becomes obvious?

• "有没有过看起来都正常，但你就是觉得哪里别扭的时候？"

1. Implicit Context & Assumed Conditions

What does the user take for granted that you might not share?

• "你做这个判断的时候，默认了哪些前提？"

1. Boundary Sense & Tolerance

Where are the fuzzy lines between acceptable and unacceptable?

• "这个程度的偏差你可以接受吗？什么时候绝对不行？"

1. Recovery & Adaptation Patterns

When things deviate from ideal, how does the user adjust?

• "如果现实条件不完美，你的默认应对方式是什么？"

1. Comparative Judgment Criteria

How does the user differentiate between options?

• "A 和 B 之间，你优先看哪个维度？"

Phase 3: Expression Barrier Breakthrough

When the user says "说不清" or gives only vague abstractions, use these techniques one at a time, starting with the gentlest:

1. Analogical Bridging

Connect to a familiar sensory or experiential domain.

• "这种感觉有点像……？比如开车、做饭、听音乐时的某种状态？"

1. Contrast Method

Surface criteria by comparing acceptable and unacceptable outcomes.

• "刚才那个'不对'的版本，和'对'的版本，最明显的差别在哪？"

1. Situational Recall

Anchor in a specific recent event rather than general principles.

• "不用总结规律，就说说最近一次——当时你注意到了什么？"

1. Elimination Method

Define what is NOT right when the user can't define what IS right.

• "如果不知道对的怎么做，那最应该避免的做法有哪些？"

Phase 4: Comprehension Model & Continuity Record

1. Summarize your understanding. Present it to the user for validation.
2. Incorporate corrections. Iterate until the user confirms: "对，就是这个意思。"
3. Record a continuity note. Store a concise comprehension model for future sessions.

Output Format: Comprehension Model

This is a lightweight calibration reference for future AI-user interactions.

# Comprehension Model: [Topic/Domain]

## Surface Statement
The user's explicit request or description.

## Deep Intent
What the user actually wants or is trying to achieve.

## Judgment Anchors
- Sensory cues the user relies on
- Timing/boundary signals
- "Right feeling" indicators

## Implicit Context
Conditions the user assumes without stating
Domain-specific defaults the user expects you to know

## Preference Mapping
- High-priority dimensions (must get right)
- Low-priority dimensions (flexible)
- Absolute no-gos

## Calibration Notes
How to recognize similar situations in future conversations

## Uncertainty Zones
What you still don't fully understand — flag for future probing

Self-Check

Before concluding, verify:

• [ ] Did I ask permission before probing deeper?
• [ ] Did I select only the most relevant 2–4 dimensions, not force all seven?
• [ ] Did I validate my understanding with the user before recording it?
• [ ] Is the comprehension model concise enough to be useful?
• [ ] Did I flag uncertainty zones rather than pretend to understand?
• [ ] Did I respect the user's choice if they wanted to stop probing?