18 KiB
Problem Map FAQ (full, beginner + practitioner)
This FAQ helps you pick the right fix fast. It is a field manual, not a brochure.
If you are new, start with “Getting started,” then jump by category.
Quick links by category
- Getting started
- Semantic gates and acceptance targets
- RAG, retrieval, and chunking
- Embeddings and vector stores
- Reasoning stability and lambda
- Memory and long context
- Agents and orchestration
- Safety and prompt integrity
- Deployment, infra, and ops
- Evaluation and governance
- Language, OCR, and PDFs
- Troubleshooting matrix
- Cost, performance, and vendor lock
- Contributing and proof
Direct map to the 16 problems
No.1 Hallucination & Chunk Drift → hallucination.md
No.2 Interpretation Collapse → retrieval-collapse.md
No.3 Long Reasoning Chains → context-drift.md
No.4 Bluffing / Overconfidence → bluffing.md
No.5 Semantic ≠ Embedding → embedding-vs-semantic.md
No.6 Logic Collapse & Recovery → logic-collapse.md
No.7 Memory Breaks Across Sessions → memory-coherence.md
No.8 Debugging is a Black Box → retrieval-traceability.md
No.9 Entropy Collapse → entropy-collapse.md
No.10 Creative Freeze → creative-freeze.md
No.11 Symbolic Collapse → symbolic-collapse.md
No.12 Philosophical Recursion → philosophical-recursion.md
No.13 Multi-Agent Chaos → Multi-Agent_Problems.md
No.14 Bootstrap Ordering → bootstrap-ordering.md
No.15 Deployment Deadlock → deployment-deadlock.md
No.16 Pre-deploy Collapse → predeploy-collapse.md
Related catalogs
Problem Map home → README.md
Semantic Clinic index → SemanticClinicIndex.md
Global Fix Map overview → GlobalFixMap/README.md
Getting started
Q. Do I need an SDK or plugin?
No. The Problem Map is text-first. Apply minimal fixes directly in your current LLM chat.
Q. I do not know my problem number. Where do I start?
Use the beginner guide and diagnose table:
- Beginner guide → BeginnerGuide.md
- Diagnose table → Diagnose.md
Q. Difference between Problem Map 1.0, Semantic Clinic, and Global Fix Map?
- Problem Map 1.0 is the 16-mode failure map with minimal fixes → ProblemMap/README.md
- Semantic Clinic is the expanded catalog by topic → SemanticClinicIndex.md
- Global Fix Map is end to end guardrails across stack → GlobalFixMap/README.md
Q. Do you have a 60-second quick start?
- One-line engine (PDF) → WFGY 1.0 Paper
- Text OS scaffold → TXTOS.txt
Semantic gates and acceptance targets
Q. What is ΔS and why gate answers with it?
ΔS is a semantic distance between question, source, and candidate answer. Gate the output under a threshold so you do not accept drift. See overview → GlobalFixMap/README.md
Q. What defaults should I enforce on every answer?
- ΔS ≤ 0.45
- Coverage ≥ 0.70
- λ state convergent across 3 paraphrases
- Source present before finalization
Q. Where do I learn how to watch these in prod?
Start from the global overview and follow the observability section → GlobalFixMap/README.md
RAG, retrieval, and chunking
Q. Model picks the wrong paragraph or mixes sources. Which number?
Usually No.1 or No.8.
- No.1 Hallucination & Chunk Drift → hallucination.md
- No.8 Debugging is a Black Box → retrieval-traceability.md
Q. What is the minimal guard for RAG answers?
Citation first, then write, with IDs and pages. See traceability → retrieval-traceability.md
Q. My chunking is brittle. Any contract to follow?
Start with the Global Fix Map overview and the Chunking section list → GlobalFixMap/README.md
Q. Retrieval is unstable after OCR PDFs.
Check OCR parsing and layout handling in the Global Fix Map overview → GlobalFixMap/README.md
Embeddings and vector stores
Q. High cosine, wrong meaning. What is that?
No.5 Semantic ≠ Embedding. Fix normalization and metric space issues. → embedding-vs-semantic.md
Q. Store-specific pitfalls to scan first?
Vector DB guardrails (per-tool)
- FAISS → faiss.md
- Chroma → chroma.md
- Qdrant → qdrant.md
- Weaviate → weaviate.md
- Milvus → milvus.md
- pgvector → pgvector.md
Q. Hybrid got worse after I mixed retrievers.
Audit metrics first, then tune weights. Guidance in the Global Fix Map overview → GlobalFixMap/README.md
Reasoning stability and lambda
Q. Coherent tone, wrong answer. Which numbers apply?
No.2 and No.6.
- No.2 Interpretation Collapse → retrieval-collapse.md
- No.6 Logic Collapse & Recovery → logic-collapse.md
Q. What is λ observe and why checkpoints?
λ state tells you if the chain is converging. Add embedded checkpoints. Primer → GlobalFixMap/README.md
Q. How do I reset without losing everything?
Use controlled reset with anchors as in No.6 → logic-collapse.md
Memory and long context
Q. Sessions forget agreements between chats.
No.7 Memory Coherence. Use state keys and guarded read/write order → memory-coherence.md
Q. Long context still drifts.
Check No.9 Entropy Collapse and add anchors → entropy-collapse.md
Agents and orchestration
Q. My agents overwrite each other’s memory.
No.13 Multi-Agent Chaos. Roles, state keys, fences → Multi-Agent_Problems.md
Q. Do these guards depend on a specific orchestrator?
No. They are provider-agnostic. See adapters in the Global Fix Map overview → GlobalFixMap/README.md
Safety and prompt integrity
Q. Prompt injection keeps leaking into the chain.
Use the Semantic Clinic entries and the safety section from the Global Fix Map overview → SemanticClinicIndex.md
Q. Users jailbreak system instructions.
Set role order and timeouts; see Global Fix Map overview → GlobalFixMap/README.md
Deployment, infra, and ops
Q. First production call fails though local tests passed.
No.16 Pre-deploy Collapse → predeploy-collapse.md
Q. Services block each other at rollout.
No.15 Deployment Deadlock → deployment-deadlock.md
Q. Canonical boot order for AI stacks.
No.14 Bootstrap Ordering → bootstrap-ordering.md
Evaluation and governance
Q. How do I prove fixes work beyond toy cases?
Use ΔS, coverage, λ convergence, and traceability. Start here → GlobalFixMap/README.md
Q. Team needs audits and sign-off.
Governance patterns live in the Global Fix Map overview → GlobalFixMap/README.md
Language, OCR, and PDFs
Q. Multilingual RAG is flaky.
Check analyzers and ranking per language. Start from the Global Fix Map overview → GlobalFixMap/README.md
Q. CJK width and punctuation break retrieval.
See the language locale notes in the overview → GlobalFixMap/README.md
Q. OCR vendor choice and parsing details.
Use the Document AI and OCR entries from the overview → GlobalFixMap/README.md
Troubleshooting matrix
Q. Output sounds right but citations are missing.
No.4 Bluffing and No.8 Traceability.
- Bluffing / Overconfidence → bluffing.md
- Retrieval Traceability → retrieval-traceability.md
Q. Every retry is different in a bad way.
No.9 Entropy Collapse and missing anchors → entropy-collapse.md
Q. Tables and equations flattened to prose.
No.11 Symbolic Collapse → symbolic-collapse.md
Q. Hybrid retrieval got worse than single model.
Audit metrics and weights from the Global Fix Map overview → GlobalFixMap/README.md
Cost, performance, and vendor lock
Q. Will guards increase token bill or latency?
Minimal guards usually reduce retries and useless generations. Measure it with your eval harness. Overview → GlobalFixMap/README.md
Q. Does this force a specific vendor?
No. The guardrails are provider-agnostic. See adapters in the overview → GlobalFixMap/README.md
Contributing and proof
Q. Where can I see real user threads or ask?
Hero Log discussions → Discussions
Q. I want to contribute modules or demos.
Start here → ProblemMap/README.md
Minimal fix checklist you can paste into any chat
- Card first. Show source IDs before writing the answer.
- Pass the semantic gate. ΔS ≤ 0.45 against the chosen source.
- Check coverage. ≥ 0.70 relative to the stated goal or query.
- Keep λ convergent across 3 paraphrases. If not, reset with anchors.
- Log the trace. Question, retrieval IDs, acceptance metrics, final answer.
🧭 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.
