Source-traceable retrieval returns every search hit with the metadata needed to find it in the original document: file path, page, position.
A retrieval result is a claim: 'this passage answers your question'. Traceability lets you check the claim. At minimum it means the original file path and page number. Better systems add the prepared-file path (the cleaned text the index was built from), a link that opens the location, and chunk metadata such as position and length. With these, a person or an agent can open the source, read the surrounding text and confirm the answer.
Traceability is the difference between retrieval and summary. Hosted memory products that store AI-extracted facts return the extraction, not the source, so there is nothing to open. Verbatim systems return the actual text with its location. For anything that matters (contracts, specifications, research, code), the second is safer. It also makes errors findable: when an answer is wrong, you can see which chunk misled the model and fix the document or the chunking.