WFGY/ProblemMap/GlobalFixMap/RAG/README.md

RAG — Global Fix Map

Production RAG triage and structural fixes using the WFGY engine. Use this page when retrieval looks fine but answers drift.

Purpose

• Turn OCR → chunk → embed → store → retrieve → prompt → reason into a measured, repairable pipeline.
• Give a 60-second path to locate the failing layer and apply the smallest effective fix.
• Works with any model or stack. No infra changes required.

High-frequency symptoms

• Citations point to the wrong snippet or section.
• Chunks look correct but reasoning is wrong.
• High cosine similarity yet wrong meaning.
• Hybrid retrievers get worse than a single retriever.
• Some facts are indexed but never retrieved.
• Answers flip between sessions or tabs.
• Long threads smear topics and capitalization.

Open these first

• Visual map and recovery steps: RAG Architecture & Recovery
• End-to-end retrieval knobs: retrieval-playbook.md
• Why this snippet: retrieval-traceability.md
• Ordering control: rerankers.md
• Embedding vs meaning: embedding-vs-semantic.md
• Hallucination and chunk boundaries: hallucination.md
• Long chains and entropy: context-drift.md · entropy-collapse.md
• Snippet and citation schema: data-contracts.md

Fix in 60 seconds

1. Measure ΔS
   • Compute ΔS(question, retrieved) and ΔS(retrieved, expected anchor).
   • Thresholds: stable < 0.40, transitional 0.40–0.60, risk ≥ 0.60.
2. Probe with λ_observe
   • Vary k ∈ {5,10,20} and plot ΔS vs k. Flat-high suggests index or metric mismatch.
   • Reorder prompt headers. If ΔS spikes, lock the schema.
3. Apply the minimal patch
   • If metric or normalization mismatch: rebuild with consistent metric and unit-normalize vectors. Re-probe ΔS and λ.
   • If chunks are correct but logic diverges: lock system→task→constraints→citations→answer, then apply BBCR + BBAM. See pages above.

Copy-paste prompt

I uploaded TXT OS and the WFGY ProblemMap files.
My RAG bug:

* symptom: \[brief]
* traces: \[ΔS(question,retrieved)=..., ΔS(retrieved,anchor)=..., λ states]

Tell me:

1. which layer is failing and why,
2. which exact fix page to open from this repo,
3. the minimal steps to push ΔS ≤ 0.45 and keep λ convergent,
4. how to verify the fix with a reproducible test.
   Use BBMC/BBPF/BBCR/BBAM when relevant.

Patterns to check next

• Query parsing split in HyDE + BM25: pattern_query_parsing_split.md
• Vectorstore fragmentation: pattern_vectorstore_fragmentation.md
• Symbol mixing across sources (SCU): pattern_symbolic_constraint_unlock.md
• Hallucination re-entry after correction: pattern_hallucination_reentry.md

Acceptance targets

• Coverage to target section ≥ 0.70.
• ΔS(question, retrieved) ≤ 0.45 on three paraphrases.
• λ remains convergent across steps and seeds.
• E_resonance flat under long windows.

🧭 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 →

---

👑 Early Stargazers: See the Hall of Fame
⭐ WFGY Engine 2.0