Ingest

Add evidence to raw/, ground it in wiki/source-notes/, then activate into synthesis (do not leave knowledge trapped in source-notes).

Steps

1. Capture — raw/inbox/ → stable raw/processed/...; never rewrite old processed meaning; add new files instead.
2. Source-note — wiki/source-notes/... with source_ids → raw paths; add Evidence summary for high-leverage sources.
3. Activate — update canonical pages, hubs, comparisons, checklists as needed.
4. Route — wiki/index.md + hubs when navigation changes.
5. Log + validate — append wiki/log.md (ingest); make validate.

PDF (text only)

Fast path when you only need page text (no figure extraction):

bash uv run python scripts/ingest_pdf.py raw/inbox/your-file.pdf

PDF, Word, PowerPoint → Markdown + images (Obsidian)

Use Pandoc when you want Markdown and embedded images on disk next to the note (Obsidian shows them via relative paths).

Install pandoc on your machine (not a Python package). PDF conversion usually also needs Poppler (e.g. brew install pandoc poppler on macOS; apt install pandoc poppler-utils on Debian/Ubuntu).

bash uv run python scripts/ingest_pandoc.py raw/inbox/report.docx uv run python scripts/ingest_pandoc.py raw/inbox/slides.pptx uv run python scripts/ingest_pandoc.py raw/inbox/scan.pdf

Output:

• raw/processed/<year>/<slug>.md — YAML frontmatter + title + body
• raw/processed/<year>/<slug>_media/ — images (…/media/… by Pandoc convention)

Open the .md in Obsidian from a vault rooted at the repo (or wiki/ — prefer repo root so raw/ paths resolve like the rest of the corpus).

If Pandoc fails on PDF, use ingest_pdf.py for text-only extraction, or fix Poppler/Pandoc and retry.

Done

Stable raw paths; source-note; activation done or deferred; routing current; log appended; validation passes.

Checklist: templates/ingest-checklist.md.

Next: wiki/index.md, then AGENTS.md § Ingest.