# Wiki Schema

## Raw Layer

- `raw/sources/<slug-documento>/` — one subfolder per source document, kebab-case slug (e.g. `iso-6336-gear-rating`, `catalogo-skf-cuscinetti-2024`). Contains the document as produced by ingestion (MinerU markdown output + `images/` + json sidecars, or a bare PDF). Never rename or move files inside a source folder — relative image links depend on the original layout.
- `raw/assets/` — standalone images/scans/figures not tied to a specific ingested document.
- Files under `raw/` are immutable once ingested. If extracted content is wrong (OCR error, bad table), fix it in the wiki page with a note — never edit the raw file.
- Every claim in `wiki/` must be traceable to a slug in `raw/sources/`.

## Page Types

| Type | Directory | Purpose |
|------|-----------|---------|
| entity | wiki/entities/ | Named things (people, tools, organizations, datasets) |
| concept | wiki/concepts/ | Ideas, techniques, phenomena, frameworks |
| source | wiki/sources/ | Papers, articles, talks, books, blog posts |
| query | wiki/queries/ | Open questions under active investigation |
| comparison | wiki/comparisons/ | Side-by-side analysis of related entities |
| synthesis | wiki/synthesis/ | Cross-cutting summaries and conclusions |
| overview | wiki/ | High-level project summary (one per project) |

## Naming Conventions

- Files: `kebab-case.md`
- Entities: match official name where possible (e.g., `openai.md`, `gpt-4.md`)
- Concepts: descriptive noun phrases (e.g., `chain-of-thought.md`)
- Sources: `author-year-slug.md` (e.g., `wei-2022-cot.md`)
- Queries: question as slug (e.g., `does-scale-improve-reasoning.md`)

## Frontmatter

All pages must include YAML frontmatter:

```yaml
---
type: entity | concept | source | query | comparison | synthesis | overview
title: Human-readable title
tags: []
related: []
created: YYYY-MM-DD
updated: YYYY-MM-DD
---
```

Source pages also include:
```yaml
authors: []
year: YYYY
url: ""
venue: ""
```

Any page that draws on a source processed via OCR/layout extraction (e.g. MinerU pipeline mode) should reference it and, where a specific fact is uncertain (garbled formula, misaligned table, ambiguous reading order), mark it inline rather than presenting it as verified:

```yaml
raw_source: raw-slug          # slug in raw/sources/ this page is derived from
confidence: high | medium | low
```

Use `confidence: low` or `medium` for any numeric value, formula, or spec pulled from a document with known extraction issues, until it's been cross-checked.

## Index Format

`wiki/index.md` lists all pages grouped by type. Each entry:
```
- [[page-slug]] — one-line description
```

## Log Format

`wiki/log.md` records activity in reverse chronological order:
```
## YYYY-MM-DD

- Action taken / finding noted
```

## Cross-referencing Rules

- Use `[[page-slug]]` syntax to link between wiki pages
- Every entity and concept should appear in `wiki/index.md`
- Queries link to the sources and concepts they draw on
- Synthesis pages cite all contributing sources via `related:`

## Contradiction Handling

When sources contradict each other:
1. Note the contradiction in the relevant concept or entity page
2. Create or update a query page to track the open question
3. Link both sources from the query page
4. Resolve in a synthesis page once sufficient evidence exists