Pack comparison
Two packs, side-by-side. Merged comparisons, shared shape, and diff highlights in one view.
rag-hybrid-bm25-vector
Sparse + dense + cross-encoder. The 2025 retrieval default.
Install
npx attrition-sh pack install rag-hybrid-bm25-vectorToken budget
8,000
Pass rate
—
Avg tokens
—
Publisher
Agent Workspace
Compatibility
Contract summary
2 required outputs, 3 permissions, 4 completion conditions.
out: outputs/retrieval.json
nine-context-sources
CLAUDE.md is user context, not system prompt. 9 sources, 4 hierarchy levels, zero embeddings.
Install
npx attrition-sh pack install nine-context-sourcesToken budget
—
Pass rate
—
Avg tokens
—
Publisher
Agent Workspace
Compatibility
Shared shape
What both packs have in common
Overlap across canonical pattern, compatibility, tags, and required packs.
Compatibility
Merged comparisons
Head-to-head claims from both packs
Each row is attributed to the pack that authored it. The winner column is normalised to this compare view (A / B / Tie).
| Source | Alternative | Axis | Winner | Note |
|---|---|---|---|---|
| A | pure-vector-rag | accuracy | A | Hybrid+rerank beats pure vector by ~10–20% NDCG@10 on technical corpora with proper nouns and code symbols. |
| A | pure-vector-rag | latency | Alternative | Pure vector is ~50ms faster — skip hybrid if latency budget is <80ms and accuracy is already acceptable. |
| A | pure-bm25 | accuracy | A | BM25 alone loses on paraphrase and conceptual queries. Hybrid adds ~15 points recall on natural-language question sets. |
| A | pure-bm25 | cost | Alternative | BM25-only has no embedding or rerank cost. Hybrid adds ~$0.001/query — negligible for most products but not all. |
| B | claude-code-guide | maintainability | Tie | Claude Code Guide is the operator recipe (how to write AGENTS.md/CLAUDE.md); this pack is the architectural why (what CLAUDE.md is and is not). Use the guide to write the file, this pack to decide what belongs in it. |
| B | rag-hybrid-bm25-vector | complexity | B | Vector-DB RAG is the canonical retrieval pattern; Claude Code's file-based memory is the opposite bet (no embeddings, LLM scan of up to 5 file headers). Compare costs: vector adds index + latency + opacity; file-based is simple and auditable. Start with file-based; measure before adopting vector. |
| B | seven-safety-layers | accuracy | Alternative | Safety layers are where enforcement lives deterministically; CLAUDE.md is where instruction lives probabilistically. For hard rules, safety-layers wins on accuracy because it actually blocks calls; this pack explains why CLAUDE.md can't. |
Diff highlights
What each pack brings that the other doesn't
Unique coverage and any measurable gap between the two.
Unique to A — Hybrid RAG: BM25 + Vector + Rerank
Comparisons not in B
Compatibility A-only
Tags A-only
Unique to B — Nine Context Sources
Comparisons not in A
Compatibility B-only
Tags B-only
Swap / reset