Lab: Build Hollin's chunker and measure it

Module 3 · Lab · About 70 minutes · Needs: lessons 1 to 4 of this module, and Module 2's lab (or the records cell in Step 1)

What you are building

Module 2 ended with out/records.jsonl: 17 typed records, 16 to chunk. This lab adds hollin/chunk/, which turns each record into chunks that later modules embed, filter and cite. Each chunk keeps its headings as a header path, carries its record's access and status fields, and gets an ID made from its content, so re-running on unchanged text changes nothing. Then you measure recall@3 for three chunking choices on 15 labelled questions with the local BAAI/bge-small-en-v1.5 model. The result is a command, python -m hollin.chunk, and tests that fail if a chunk loses text, mixes documents or orphans a table row. No model key is needed.