every question returns the same three chunks after i moved to 1500 token chunks chunking

Bigger chunks average the signal away. One 1500-token chunk covers six topics, so its vector sits near the middle of everything, which makes it a decent match for every query and a good match for none. The bunched distance range is the fingerprint of exactly that — your scores have stopped carrying information.

What I would do:

• Embed small, return big. Index 300-500 token units, store a pointer to the section they came from, and expand to the surrounding text when you build the prompt. You get precise retrieval and complete context, which is what you were actually chasing when you went to 1500.
• Strip boilerplate before embedding. Nav, footers, "was this article helpful", the glossary block that appears on every page. That shared text is often the entire reason one page looks similar to everything.

Small-to-big, two columns, one join. chunk(id, doc_id, section_id, text, embedding) and section(id, text). Search chunks, group hits by section, pass sections to the model. It is fifteen minutes of work and it is the single highest-leverage change in most naive pipelines.

0.31 to 0.38 for everything means you cannot rank on distance any more and no amount of threshold tuning will fix it. Put a cross-encoder reranker on the top 50 — it reads the query and the chunk together rather than comparing two independently produced vectors, and it will separate what cosine has flattened.

Check for duplicated content generally. If a paragraph is present in 200 pages, all 200 look alike to the index and one of them wins arbitrarily. A quick group-by on a hash of the chunk text usually turns up something ugly.

Also worth checking what you are embedding on the query side. If you pass the raw user message including three paragraphs of preamble and a pasted stack trace, the query vector is as smeared as your fat chunks are. Cleaning up the query moved our recall more than chunk size did.

MMR on the results will diversify them and push the duplicate onboarding page down.