# source/ — AI-Cleaned Source Documents

One markdown file per source document, produced from the corresponding file(s)
in `raw/`. These are the direct input to wiki ingest.

Conventions:

- Organize by document type: `source/<doc-type>/<slug>.md` (create subfolders
  as your domain requires — mirror the `doc-type:` values in CLAUDE.md).
- Every file starts with a provenance comment before the title, naming the
  extraction tool, e.g. `<!-- source: pdftotext -->` or
  `<!-- source: OCR (tesseract) -->`.
- Cleaning means fixing extraction artifacts and restoring structure
  (headings, lists, tables). **Never add or interpret content** — that happens
  in the wiki layer, with citations.
- Separate major chapters/parts with a `---` horizontal rule on its own line,
  followed by a blank line and the chapter heading, so chapter boundaries can
  be located programmatically: `grep -n "^---$" source/<type>/<slug>.md`