Blog (draft): Hybrid search ranking — Part 1 of 2

[g-despot]

Part 1 of a 2-part "Search result quality & ordering" series

A practical, runnable tour of Weaviate's query-time ranking dials, threaded through a documentation / RAG DocChunk example (chunked pages, near-duplicates, stale versions — "relevant-ish but wrong order"):

• Hybrid alpha — the vector-vs-keyword weight (default 0.75), with an alpha-sweep snippet + a compact BM25 sidebar (per-property AND gotcha).
• Fusion — relativeScoreFusion (the server default) vs. rankedFusion (RRF, k=60 hardcoded); when to pick which.
• Reranking — the six reranker modules; the key teaching that a reranker reorders only the retrieved page, so over-fetch first.
• Boost API — soft, never-excludes rescoring; blend a recency decay + current-version lift.
• A combined hybrid → rerank → boost query and a symptom → tool decision table. Teases Part 2.

~2,050 words. Every snippet was checked against the current Python client; the alpha-sweep, both fusion modes, and the Boost call ran against a local Weaviate 1.38.

Open before publish (tracked as inline TODOs — opening as draft)

• Hero image (./img/hero.png) + restore the image: frontmatter line
• Boost API GA status — Preview in 1.38 today; confirm it's GA and stamp the exact version, or reframe the section with a preview caveat
• Set the real publish date (currently a placeholder)
• Link Part 2 ("Clean result sets: dedupe, diversify, and explain") once it publishes
• Final human review

Part 2 to follow as a separate PR.

[orca-security-eu]

Orca Security Scan Summary

Status	Check	Issues by priority
Passed	Infrastructure as Code	0   0   0   0	View in Orca
Passed	SAST	0   0   0   0	View in Orca
Passed	Secrets	0   0   0   0	View in Orca
Passed	Vulnerabilities	0   0   0   0	View in Orca

[g-despot]

Revised after review (2nd commit). Changes: concrete RAG failure as the opening hook; question/keyword-framed headings; three Mermaid diagrams (retrieval→ranking→presentation pipeline, hybrid fusion, over-fetch→rerank→boost flow); inline definitions for jargon (cross-encoder, bi-encoder, RRF, min-max normalization, over-fetch); an faq block (FAQPage JSON-LD); primary keyword "hybrid search ranking" in title/description/slug/headings. Code snippets and all load-bearing facts are unchanged; Boost GA/version TODOs preserved. Still a draft — see the open checklist above (hero image, Boost GA confirmation, publish date, Part 2 link).