An architect is designing a RAG system for a 10,000-document legal corpus where users ask highly specific questions requiring precise clause-level retrieval. Which chunking strategy is most appropriate?

Hierarchical/semantic chunking is optimal for legal documents because legal meaning is inherently hierarchical: a clause only makes sense in the context of its section, which only makes sense in the context of the document's structure. Parent-child chunking enables precision retrieval (child chunks for specific clauses) while maintaining context (parent chunks provide surrounding structure). This addresses both precision and context needs. Option A (document-level) retrieves too much irrelevant content per chunk, overwhelming Claude's context and diluting relevance signals. Option B (fixed 512-char) cuts across legal structure arbitrarily, splitting clauses mid-sentence and destroying legal meaning at boundaries. Option D (sentence-level) maximizes precision but loses the surrounding context that makes individual sentences legally meaningful — sentences like 'This obligation shall not apply' are meaningless without their preceding clause.