insight-externalizer

Insight Externalizer

Turn tacit user understanding into artifacts that can be inspected, corrected, and used. Protect the user's view: do not substitute the model's understanding for the user's understanding.

Use after intent framing and before argument spine work when the point of view, causal model, assumptions, or key distinction is still implicit.

Inputs

• Approved intent brief or enough context to preserve audience and desired change.
• User notes, transcripts, source excerpts, examples, anecdotes, data, or rough claims.
• The user's own explanation of what matters, what is misunderstood, and what they believe others miss.
• Any known disputes, uncertainties, or missing evidence.

Outputs

Return an Insight Artifact with:

• plain_language_explanation: the user's idea in direct, non-jargony language.
• concept_map: key concepts and relationships, including cause, contrast, dependency, sequence, and tradeoff.
• assumptions_ledger: assumptions, why they matter, confidence, and how to verify them.
• fact_inference_assumption_labels: labeled statements so claims are not blurred together.
• open_questions: questions that must be answered before argument or slide work can proceed confidently.

Use this compact format:

## Insight Artifact
- Plain-language explanation:
- Concept map:
- Assumptions ledger:
- Fact / inference / assumption labels:
- Open questions:
- User confirmation needed:

Workflow

1. Restate the user's view using their terms first, then simplify only after preserving meaning.
2. Extract key concepts, actors, forces, metrics, tradeoffs, and tensions.
3. Build a compact concept map in text using arrows or grouped bullets.
4. Label each important statement as fact, inference, or assumption.
5. Create an assumptions ledger for claims that carry persuasive or factual risk.
6. Compare the emerging insight against the intent brief: ask whether it helps move the primary audience from current state to desired change.
7. Ask targeted questions where the user's view is missing, not where the model merely wants more detail.
8. Return the artifact and ask the user to correct it before argument spine work.