Mission

This tree exists so an AI agent (or a curious human) can navigate FY-2026-27-Proposed-Budget.pdf without loading the whole PDF into a model. Raw PDFs of this size cost 100K+ tokens just to read; a well-indexed tree lets a question-answerer traverse only the sub-tree it needs, usually a few KB.

Navigation contract

1. Start at INDEX.md. It is kept short (< 200 lines).
2. Every INDEX.md lists its children with page ranges and direct PDF deep-links.
3. Every leaf note cites its source PDF pages in the body and in front matter.
4. To verify any claim: click the "PDF pp. N-M" link to open the original PDF at that page.
5. To see the raw extracted text a leaf was built from, click the "raw: 412 · 413 · ..." page links.

File kinds

• INDEX.md -- a map, not content. Short, scannable, always links to children.
• <name>.md -- a leaf section holding raw extracted text plus source citations.
• <name>-part-N.md -- a chunk of a long leaf, split so each file stays under ~8K tokens.

Provenance

All content is derived mechanically from the PDF. No human has rewritten text. Stage 5 enrichment (if present) adds a visually distinct ## TL;DR block with per-bullet page citations; raw text below is untouched.