Add KB Arena retrieval benchmark notebook

[xmpuspus]

Adds a new cookbook notebook: benchmark_retrieval_strategies_kb_arena.ipynb.

What it shows

Before wiring a retrieval approach into a Haystack pipeline, the notebook walks through using KB Arena to compare retrieval architectures on a small example corpus and then map the winning strategy to Haystack's InMemoryBM25Retriever.

KB Arena benchmarks nine architecturally distinct retrieval strategies head-to-head and reports IR metrics with paired-bootstrap 95% CIs and Wilcoxon p-values:

• naive vector, contextual vector, QnA pairs, knowledge graph (Neo4j), hybrid RRF, RAPTOR, PageIndex, BM25, rerank-vector (cross-encoder)
• Recall@k, Precision@k, MRR, NDCG (binary + graded), MAP, R-Precision, bpref, RBO
• Pareto frontier across (NDCG, latency)

What's in the PR

• notebooks/benchmark_retrieval_strategies_kb_arena.ipynb — 18 cells covering install, corpus prep, ingest, retriever-lab, results inspection, and Haystack pipeline wiring. Runs end-to-end with no API keys (BM25-only path).
• index.toml — new [[cookbook]] entry tagged ["Evaluation", "Advanced Retrieval", "RAG"] with new = true.

Notes

• Uses a tiny inline corpus (three AWS-flavored Markdown files) so the notebook is self-contained and fast.
• The full nine-strategy benchmark requires Neo4j and an embedding provider; the notebook calls those out as next steps rather than requiring them.
• KB Arena: MIT, on PyPI as kb-arena, Zenodo concept DOI 10.5281/zenodo.20319678.

Filename convention matches the existing pattern (descriptive, includes the technologies involved). Follows the cookbook README's three contribution rules: notebook in /notebooks, descriptive filename, index.toml entry with title and topics.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB