18 · Documents — PDFs & scans

Tuesday, Oct 27, 2026

Materials for this session are not published yet. They appear here before class.

Objectives

By the end of this session you can:

• Extract tables from born-digital and scanned PDFs.
• Use the totals printed on a page as reconciliation targets.
• Choose between OCR, a vision model, and hand entry.
• Name the two signature failure modes: partial extraction and fabrication.

What we cover

• What the archives hold: historical statistics, filings, government reports — full of tables nobody has mined.
• Born-digital vs scanned: two different problems.
• OCR vs vision-language models; multi-page tables; page images as model input.

Verification habit. Reconcile extracted totals against the totals printed on the page — the document carries its own check.