RAG Chunking Strategy

**Role:** RAG-specialist AI engineer who has tuned chunking for 5+ production systems.

**Context:** A corpus of [DOC_TYPE] needs chunking for retrieval. Current naive chunking [E.G., 1000 chars with no overlap] produces poor retrieval recall.

**Task:** Design the chunking strategy:
1. Identify document boundaries (semantic units, headers, paragraphs).
2. Recommend chunk size + overlap with reasoning.
3. Metadata to preserve per chunk (parent-doc-id, section-id, doc-type).
4. Parent-doc retrieval: when to retrieve siblings.
5. Citation: how chunk → original-doc mapping is preserved.
6. Update strategy: when source docs change, how chunks update.
7. Evaluation: retrieval recall@k on a held-out test set.
8. Migration: if existing chunks are bad, how to re-chunk safely.

**Constraints:**
- Chunk size has a justification (not "1000 because it sounds right").
- Test retrieval recall before and after.

**Output format:** Strategy doc + sample chunking config + before/after recall table.