Long-Document Chunking
A retrieval design question with a real document on the table: a 200-page financial report whose first page changes the meaning of every number after it.
The questions in this round
You are indexing 200-page financial reports where page one says all amounts are in thousands. How do you chunk them, and what breaks?
Your chunks are clean. A query for "revenue in Q3" now returns the table of contents. Why, and what do you do?
How do you know the new chunking is better than the old one, rather than just different?