A civic-arboretum FAQ is being loaded into a vector store. One pass cuts every 2,000 characters with no overlap, and questions that straddle a cut come back empty. Which ingest knobs matter?
Select an answer to reveal the explanation.
Short Explanation
Every 2,000 characters with no overlap drops questions that straddle a cut. Chunk size and overlap are ingest-quality knobs that change retrieval. They are not the RAG definition, a client query, or a Guardrails policy.
Full Explanation
Chunk size and overlap are data-prep decisions for RAG ingest. A hard 2,000-character cut with no overlap drops questions that straddle a boundary. This is not the retrieve-then-generate definition and not a client query. Adjust the ingest knobs so retrieval can see the straddling text.