Ridgeline's Data 360 admin sets an unusually large chunk size when indexing the clinic policy manual. What is the most likely downside during agent grounding?
Select an answer to reveal the explanation.
Short Explanation
Oversized chunks are like handing someone the whole chapter when they asked what page the rule is on — the answer's in there, but it's buried in noise the model has to wade through, which hurts precision.
Full Explanation
Chunk size is a real trade-off: chunks that are too large tend to bundle the relevant sentence together with unrelated policy text, which dilutes retrieval precision and can push the model toward vaguer or less accurate answers because it has to sift signal from noise. Chunks that are too small lose surrounding context and can fragment an idea across chunk boundaries. There is no indication that Salesforce enforces a hard one-paragraph maximum that would cause an outright indexing failure — that overstates the constraint into something the docs don't state. The document doesn't become entirely unreferenceable; it still indexes and retrieves, just less precisely, so 'unable to reference at all' is too extreme. Chunk size also isn't what determines whether keyword or vector search is used — that's a separate, deliberate configuration choice on the retriever, not a side effect of chunk sizing. The exam-relevant takeaway is that chunking strategy is a tunable lever affecting retrieval precision, not a binary pass/fail switch.