---
name: structuring-knowledge
description: Use when raw or elicited material must become structured knowledge (concepts, claims, examples, contradictions, open questions), or when the load-bearing claims of an argument need mapping, or when a claim graph has orphans or circular support
---

# Structuring knowledge

## Overview

The knowledge base is the author's structured knowledge: concepts, claims, examples, contradictions, open questions, and the claim graph that holds an argument together. It is shared across every work in the workspace.

This skill turns the loose candidates from `ingesting-sources` and `interviewing-the-author` into a structure that later passes read from. It also carries two knowledge-band disciplines the manifesto promotes to first class: the argument spine and the claims index.

## The disciplined move

Structure is not summary. The job is to expose the architecture of what the author knows, including where it is weak, not to produce a tidy précis that hides the seams.

**Concept and claim capture.** Every load-bearing claim gets an entry. Each claim records what supports it and what it depends on.

**Argument spine.** Build the claim graph for the active work. Surface orphan claims (nothing supports them) and circular support (A leans on B leans on A). Report these; do not quietly patch them.

**Claims index.** Maintain `claims.md`: every load-bearing claim mapped to at least one source in the sources. A claim with no source is flagged, never silently accepted.

**Gap-marking, not confabulation.** When the structure needs a fact, figure, anecdote, or citation the author has not supplied, insert `[AUTHOR: needed — what]` and stop. The system is structurally unable to pretend to know what only the author knows.

## Rationalization table

| What you will tell yourself | The reality | Do instead |
|---|---|---|
| "A clean summary is good structure." | Summary smooths over the gaps that the author needs to see. | Map claims and dependencies; expose orphans and circular support. |
| "This claim is common knowledge, no source needed." | "Common knowledge" is where unsupported assertions hide. | Add it to the claims index flagged as unsourced until the author supplies one. |
| "I can supply the obvious citation." | A model-supplied citation is untrusted until verified against a real index. | Mark `[AUTHOR: needed — citation]`; verification belongs to `auditing-adversarially`. |
| "These two claims roughly agree, I will merge them." | Merging can hide a real contradiction the author should keep. | Keep both; log the tension in contradictions. |

## Red flags: stop if you catch yourself

- Writing a summary in place of a claim graph.
- Accepting a load-bearing claim with no source.
- Inventing a citation, DOI, page number, or statistic.
- Erasing a contradiction to make the structure look finished.

## Output shape

```markdown
## Knowledge update
### Concepts
### Claims (id → statement → support → depends-on → source?)
### Argument spine (claim graph; orphans and cycles flagged)
### Examples
### Contradictions
### Open questions
### Gaps marked: [AUTHOR: needed — ...]
```

## Acceptance and edge cases

- **Legitimate co-definition vs vicious circularity:** two claims that mutually define each other are not always a defect. Distinguish a valid definitional loop from circular *support*, and present rather than condemn.
- **A claim the author calls "obvious":** keep it flagged as unsourced regardless of their confidence; obviousness is not a source.
- **A "claim" that is actually a value or preference** (not falsifiable): classify it as stance, not claim, and do not demand evidence for a taste.
- **A whole book's worth of claims:** scope the claim graph to the active unit each run; a corpus-wide graph is unreadable and unusable.

## Next

Voice material goes to `managing-voice`. A structured knowledge base plus a selected register unlocks `composing-as-amanuensis`. Source verification of the claims index happens in `auditing-adversarially`.