Text extraction and pagination

For the agent to read and cite a document, you extract its text, and how you extract it matters for citation precision (Chapter 7). Annotate the extracted text with stable location markers (page numbers, section ids) that line up with what the user sees in the rendered view. Then a citation that says "page 3" can scroll the viewer to page 3 and highlight the quote. Don't trust page numbers printed inside the document (footers, roman numerals); use your own sequential markers tied to the rendering. Extraction and rendering must agree on locations, or citations point to the wrong place.