I am seeing a recurring theme here. RAG has a dependency on a high quality search solution that is already in place and providing value to users. Anything else

I am seeing a recurring theme here. RAG has a dependency on a high quality search solution that is already in place and providing value to users. Anything else now looks like putting the cart before the horse. If you cannot get a basic search term to return documents you care about, I do not think going further into the vector/semantic rabbit hole will make your life much better. At this point I think you need to step back and ask why your simple terms aren't doing what you would expect and start working on that problem. Perhaps there is something wrong with your documents or business rather than the technology. With regard to traceability, if you have a traditional search foundation to work with, presumably you have fields like "IndexedUtc" and "Url" for every document that has been indexed. This metadata could be concatenated to any generative output. These additional metadata fields are also how you can tune the search process from a RAG perspective. During indexing, you can record heuristics about each document to condition how the text fragments are extracted and transformed into the RAG prompt.