8.4 KiB
Metric Mismatch — Guardrails and Fix Pattern
🧭 Quick Return to Map
You are in a sub-page of RAG_VectorDB.
To reorient, go back here:
- RAG_VectorDB — vector databases for retrieval and grounding
- WFGY Global Fix Map — main Emergency Room, 300+ structured fixes
- WFGY Problem Map 1.0 — 16 reproducible failure modes
Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.
Use this page when nearest neighbors look similar in cosine space but your VectorDB runs L2 or dot, or the reverse.
This failure appears often in FAISS, Milvus, pgvector, Weaviate, Redis, Vespa, and similar stores.
Open these first
- Visual map and recovery: RAG Architecture & Recovery
- End-to-end retrieval knobs: Retrieval Playbook
- Embedding vs meaning: embedding-vs-semantic.md
- Chunking checklist: chunking-checklist.md
Core acceptance
- ΔS(question, retrieved) ≤ 0.45
- Coverage ≥ 0.70 for the target section
- λ remains convergent across three paraphrases and two seeds
- Store metric matches embedding training metric (cosine ↔ cosine, L2 ↔ L2, dot ↔ dot)
Typical breakpoints and the right fix
-
High cosine similarity in logs but wrong meaning
→ embedding-vs-semantic.md -
Top-k neighbors inconsistent across runs (vector drift between L2 and cosine)
→ retrieval-playbook.md -
Switching embedding models breaks index (new default metric not aligned with store)
→ predeploy-collapse.md -
Hybrid dense+BM25 loses semantic signal (wrong weighting due to metric scaling)
→ hybrid_failure.md
Store defaults reference
| Store | Default metric | Notes |
|---|---|---|
| FAISS | L2 (can set IP or cosine) | Normalize vectors before cosine search |
| Milvus | L2 / IP | Cosine requires explicit normalization |
| pgvector | L2 / cosine / IP | Must choose at index creation |
| Weaviate | cosine | Dot/IP optional |
| Redis-Vector | cosine | Normalize mandatory |
| Vespa | dot | Needs scaling to emulate cosine |
Fix in 60 seconds
-
Log current metric
Run a probe query (SELECT metric FROM index_metadata). Verify it matches embedding doc. -
Check normalization
If metric=cosine but vectors are raw, ΔS will inflate. Normalize to unit length. -
Re-index with explicit metric
Drop and rebuild index with the same metric as embedding training. -
Hybrid sanity check
If using BM25+dense, reweight so ΔS ≤ 0.45 and coverage ≥ 0.70.
Copy-paste test query
-- Example: pgvector
SELECT id, embedding <=> query_embedding
FROM documents
ORDER BY embedding <=> query_embedding
LIMIT 5;
Ensure <=> operator matches the chosen metric (cosine, L2, or IP).
🔗 Quick-Start Downloads (60 sec)
| Tool | Link | 3-Step Setup |
|---|---|---|
| WFGY 1.0 PDF | Engine Paper | 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>” |
| TXT OS (plain-text OS) | TXTOS.txt | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
🧭 Explore More
| Module | Description | Link |
|---|---|---|
| WFGY Core | WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack | View → |
| Problem Map 1.0 | Initial 16-mode diagnostic and symbolic fix framework | View → |
| Problem Map 2.0 | RAG-focused failure tree, modular fixes, and pipelines | View → |
| Semantic Clinic Index | Expanded failure catalog: prompt injection, memory bugs, logic drift | View → |
| Semantic Blueprint | Layer-based symbolic reasoning & semantic modulations | View → |
| Benchmark vs GPT-5 | Stress test GPT-5 with full WFGY reasoning suite | View → |
| 🧙♂️ Starter Village 🏡 | New here? Lost in symbols? Click here and let the wizard guide you through | Start → |
👑 Early Stargazers: See the Hall of Fame — Engineers, hackers, and open source builders who supported WFGY from day one.
⭐ WFGY Engine 2.0 is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the Unlock Board.
