wiki-ingest

# Wiki Ingest

Ingest a source document into the knowledge wiki — creating structured, interlinked pages that compound over time.

## Usage Template

**Prompt**
```text
Use wiki-ingest on this source. Create source notes, concept pages, entity pages, navigation updates, and a verification summary.
```

**Use Case**
- Turning an article, PDF, transcript, or rough note into durable linked knowledge.

**Expected Result**
- The agent creates an immutable source note, 3-7 key insights, linked wiki pages, and a log entry.

**Output Example**
- `SOURCES_DIR/src-YYYYMMDD-title.md`, `CONCEPTS_DIR/concept-name.md`, `ENTITIES_DIR/entity-name.md`, and `LOG_FILE` update.

**Verification Case**
- New wiki pages have frontmatter, source references, at least two `[[wikilinks]]`, and timeline entries.

**Verified Effect**
- Knowledge stops being a loose summary: the source becomes traceable, linked, and reusable across future sessions.

## Success Metrics

- Creates one immutable source note, 3-7 key insights with block refs, and at least one linked wiki page.
- Each new wiki page has frontmatter, source references, at least two wikilinks, and a timeline entry.
- Final report lists created/updated files and flags single-source claims.
- Targeted post-ingest lint confirms no missing frontmatter, broken source refs, zero-inlink pages, or pages with fewer than two outbound wikilinks.
- Clippings are archived after successful ingest and the clipping queue is updated when applicable.
- Each concept page passes the Karpathy understanding gate: one-line thesis, source boundary, mechanism, connections, and what remains uncertain.

**V5.2 Closure Add-on**
- Classify every input as `external-fact`, `human-experience`, `internal-state`, or `environment-signal`.
- For high-value ingests, decide whether to create one behavior experiment and one creativity experiment.
- Record governance risks: single-source claims, missing provenance, stale-page risk, and review queue items.

## When to Use

- User drops a file and says "ingest this"
- User shares a link/article/video/PDF to be processed
- User says "add this to the wiki" or "save this to my knowledge base"
- User mentions a source that should be captured

## Workflow

### Step 0: Resolve Paths

Before writing, read `system/config.md` when available and resolve `SOURCES_DIR`, `CONCEPTS_DIR`, `ENTITIES_DIR`, `MAPS_DIR`, `LOG_FILE`, `BEHAVIORS_DIR`, and `CREATIVITY_DIR`. If no config exists, use the default STOW layout.

### Step 0A: Define the Ingest Macro Action

Treat each ingest as an agentic macro action:

```text
Objective:
Source boundary:
Owned vault paths:
Expected source note:
Expected wiki updates:
Verification evidence:
Non-goals:
Stop condition:
```

Do not start a broad autonomous expansion unless the source has objective metrics and cheap verification. Otherwise ingest once, record uncertainty, and queue follow-up research.

### Step 1: Capture

Create an immutable source note in `SOURCES_DIR`:

```markdown
---
source_id: "src-YYYYMMDD-short-slug"
source_date: "YYYY-MM-DD"
source_title: "Original Source Title"
source_author: ""
source_type: "article | book | video-transcript | pdf | epub | notebooklm-mediated | local-synthesis | primary-filing | company-interview"
source_url: ""
input_class: "external-fact | human-experience | internal-state | environment-signal"
created: "YYYY-MM-DD"
knowledge_stage: captured
evidence_level: "single-source | multi-source | curated-map"
trust_level: "1-unverified | 2-expert-source | 3-primary-source"
hash: "sha256-16char"
status: "raw | ingested"
---
```

**Rules:**
- Never modify source files after creation
- Extract 3-7 Key Insights with block refs (`^ki-short-name`)
- Flag single-source claims with `> [!warning] Single source`
- Assign an input class: `external-fact | human-experience | internal-state | environment-signal`

### Step 1A: Classify Source Risk

Use a source-risk label before writing claims:

| Source type | Default evidence | Required caution |
|---|---|---|
| primary filing / official docs | high | still check date, draft/final status, and missing fields |
| article / book / transcript | medium | mark author perspective and single-source claims |
| founder / company / investor interview | medium-low | treat metrics, adoption, roadmap, and performance as self-reported |
| NotebookLM-mediated source | low | state that the original transcript/source is not fully archived |
| local synthesis / PDF stack | low | treat as secondary-source synthesis; do not promote numbers without primary refs |
| fast-changing financial/product claim | low | add review queue item for current docs or filings |

If a clipping duplicates an existing source, create or update provenance only when it adds useful block refs; otherwise link to the canonical source and log the duplicate.

### Step 2: Auto-create Entity Pages

For every person, company, product, or project mentioned in the source:

```markdown
---
tags:
  - domain/strategy
  - type/entity
type: entity
status: seed
created: "YYYY-MM-DD"
knowledge_stage: single-source
evidence_level: single-source
---

# Entity Name
```

- Check if entity page already exists → update if needed
- Use consistent naming (English for global entities, Chinese for local)

### Step 3: Create/Update Concept Pages

For every key concept, framework, or methodology:

- Check if concept page already exists
- If exists: add new source reference, flag contradictions
- If new: create with `status: seed`, ≥2 `[[wikilinks]]`, `knowledge_stage: single-source`

**Standardized page structure** (all concept pages should follow this pattern):

```markdown
# Title

> Core thesis in one line — what is the single most important thing to know?
> (Source: [[source]])

---

## Core Mechanism

ASCII diagram showing causal relationships.

## Classifications / Comparisons

Tables comparing with related frameworks.

## Key Data (if applicable)

Bulleted list of critical numbers.

## Implications / Applications

What t