Because of token limits, and chunking, one section can be represented many times in the docs, which can lead to it being "selected twice" in the retrieval, with different scores (this is NOT a bug). However, it would be nice to only keep one of those retrieved sources, or clearly distinguish between them. See screenshot:
Because of token limits, and chunking, one section can be represented many times in the docs, which can lead to it being "selected twice" in the retrieval, with different scores (this is NOT a bug). However, it would be nice to only keep one of those retrieved sources, or clearly distinguish between them. See screenshot: