fix(docs): forward top_n to rerank in gen-qa-openai notebook

[RaphaelFakhri]

Problem

In docs/gen-qa-openai.ipynb, rerank(query, results, top_n=5) ignores its top_n argument and always sends top_n=10 to pc.inference.rerank. As a result:

• cascading_retrieval(..., top_n=5) in the "Cascading Retrieval" section returns 10 papers instead of 5.
• rag(query) asks for top_n=5, but the prompt sent to gpt-4o-mini contains 10 papers, which doubles the context size and token cost and contradicts the notebook text ("returning the top_n results").

Solution

Pass the top_n argument through to pc.inference.rerank. This is a one-line change.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Test plan

The repository has no unit test harness for notebooks, so the cell that defines query_pinecone, dedupe, rerank and cascading_retrieval was executed against stub pc and index objects that record the arguments sent to pc.inference.rerank:

• Before the change: top_n=3 and top_n=5 both send top_n=10 and return 10 results (0 passed, 2 failed).
• After the change: top_n=3 returns 3 results and top_n=5 returns 5 results (2 passed, 0 failed).

ruff check, ruff format --check, codespell and the four notebook checks from the lint workflow (check-secrets, check-structure, check-timeless, check-pinning) pass on the changed notebook.

---

Note

Low Risk
Documentation-only notebook fix with no production code or security impact.

Overview
Fixes a bug in the gen-qa-openai notebook where rerank() accepted a top_n argument but always called pc.inference.rerank with a hardcoded top_n=10.

The call now passes through the function’s top_n parameter, so cascading_retrieval and rag honor values like top_n=5 instead of returning ten papers and inflating the RAG prompt sent to gpt-4o-mini.

^{Reviewed by Cursor Bugbot for commit 6d03772. Bugbot is set up for automated code reviews on this repo. Configure here.}