kb-search Enhancements Proposal

1. LLM reranking stage HIGH IMPACT

Add a cross-encoder reranking pass over the top-K hybrid candidates. qmd uses qwen3-reranker-0.6b (~640MB GGUF) — small enough to sit alongside bge on the 4070 permanently. Flow: hybrid retrieval pulls ~40 candidates → reranker scores each (query, chunk) pair → final order blends retrieval and reranker scores.

qmd's position-aware blend is worth copying wholesale: rank 1–3 keep 75% retrieval weight, 4–10 get 60%, 11+ get 40%. This stops the reranker destroying exact-match hits while letting it rescue mid-ranked semantic matches.

API: add rerank: bool (default true) to the search endpoint, with --no-rerank in the CLI for latency-sensitive callers.

2. RRF fusion MEDIUM-HIGH

Replace the current score blend with Reciprocal Rank Fusion when merging FTS and vector lists: score = Σ 1/(k + rank + 1), k=60. Rank-based fusion sidesteps the incomparability of BM25 scores (unbounded) and cosine similarity (0–1). qmd adds a top-rank bonus (+0.05 for #1, +0.02 for #2–3 in any list) to preserve exact matches — cheap and effective.

Pure engine-side change, no API impact. Scores become comparable across queries too, which fixes the "score is relative, not absolute" caveat in the current skill docs.

3. Bench harness + explain traces DO FIRST

We currently have no way to know if any of the above helps. Add:

• kb bench fixture.json — run a fixture of queries with known-relevant docs, report precision@k / recall / MRR per backend (fts-only, vec-only, hybrid, hybrid+rerank). Directly mirrors qmd bench.
• --explain on search — per-result score breakdown (FTS score, vector score, fusion contribution, rerank score).

A fixture of 20–30 real queries against the existing corpus (lab infra questions, manual lookups, note recall) gives a regression baseline before touching ranking. This should land before #1 and #2 so their benefit is provable.

4. Context descriptions MEDIUM

qmd's standout idea: attach a one-line description to a collection or path (e.g. "Meeting transcripts", "Lab infrastructure runbooks") and return it with every matching result. For kb, the natural unit is the tag: kb tag-describe ops "Lab operations runbooks and procedures", returned as tag_contexts in search results. Helps an LLM consumer (me) judge which of several similar-scoring chunks actually answers the question — descriptions cost nothing at query time.

5. Query expansion LATER

qmd fine-tuned a 1.7B model to generate 2 query variants, searching all three and fusing via RRF. Real quality gains, but the heaviest lift: another model resident in VRAM, ~1–2s latency, and much of the benefit is available cheaper — I already do multi-query decomposition client-side per the kb skill. Park until #1–#3 have landed and the bench shows remaining headroom.

6. JSON ingestion TRACKED

The original scope of this task: kb rejects .json uploads, forcing renames to .txt. Add .json (and sensibly .yaml/.yml/.toml) to the accepted extensions, ingesting as text. Optional nicety: pretty-print minified JSON before chunking so chunks break on structure.