Module 1 · Reading real-world documents · Lesson 3
Tables, plans and handwriting
The three kinds of content every parser finds hardest in a real case file, how to represent each faithfully, and the checks and review routes that catch what parsers get wrong.
By the end of this lesson you can
- Choose a table representation that preserves merged cells and multi-page tables, and verify extracted tables with arithmetic invariants
- Explain what a plan or drawing can and cannot tell an agent, and cross-check its references against the text with simple computer vision
- Design handwriting and checkbox extraction with explicit illegible and ambiguous states
- Compute the review rate and residual error of a two-reader agreement check, and state its independence assumption