Long documents contain several unrelated sections. Which embedding design should be evaluated?

1. Duplicate the same vector repeatedly
2. One unbounded prompt containing every document for every request
3. Remove all document and section references
4. Meaningful chunks with source identifiers and appropriate overlap