Fixed-length chunking (e.g., 500 tokens per chunk) is low-cost to implement but tends to cut across semantic boundaries, causing fragments to lack complete context.