WFGY/ProblemMap/BeginnerGuide.md

🆕 Beginner Guide — How to Identify & Fix Your AI Failure

A zero-to-hero crash-course for anyone new to WFGY, RAG pipelines, or “why is my model hallucinating?”

If the full Problem Map feels overwhelming, start here.
In ~10 minutes you’ll locate your failure family, run a safe first fix, and know how to verify it.

---

Quick Nav
Getting Started (Practical) · Problem Map 2.0 (RAG) · Patterns Index · Examples · Eval · Ops

---

0) 🎯 Why this guide exists

When RAG breaks, it’s rarely one bug. It’s stacked illusions across OCR → chunking → embedding → retrieval → prompt → reasoning.
This guide helps you:

1. Identify the failure family fast
2. Apply the minimal structural fix (not prompt band-aids)
3. Verify with objective signals: ΔS (semantic stress), λ_observe (layered states), E_resonance (coherence)

Then jump deeper via Problem Map 2.0 and Patterns.

---

1) 🔍 “Which symptom matches my bug?”

Follow the first Yes you hit; then open that page.

Question	Yes → Open	No → Next
Chunks look correct but the answer is wrong?	hallucination.md	↓
Reached the right chunk but logic fails?	retrieval-collapse.md	↓
Multi-step tasks derail after a few hops?	context-drift.md	↓
Model gives confident nonsense?	bluffing.md	↓
High similarity scores but wrong meaning?	embedding-vs-semantic.md	↓
Logic dead-ends / loops?	logic-collapse.md	↓
Long chat forgets context?	memory-coherence.md	↓
Can’t trace why it failed?	retrieval-traceability.md	↓
Output becomes incoherent / repetitive?	entropy-collapse.md	↓
Replies turn flat / literal?	creative-freeze.md	↓
Formal/symbolic prompts break?	symbolic-collapse.md	↓
Paradox/self-reference crashes?	philosophical-recursion.md	↓
Multi-agent roles/memory collide?	multi-agent-chaos.md	↓
Tools fire before index/data ready?	bootstrap-ordering.md	↓
Services wait on each other forever?	deployment-deadlock.md	↓
First prod call crashes after deploy?	predeploy-collapse.md	File an Issue →

Extended patterns (very common in the wild):

• Hybrid HyDE+BM25 gets worse than single → patterns/pattern_query_parsing_split.md
• Two sources merge into one (who-said-what mixes) → patterns/pattern_symbolic_constraint_unlock.md
• You correct it, bad fact returns later → patterns/pattern_hallucination_reentry.md
• State flips across tabs/sessions → patterns/pattern_memory_desync.md
• Some facts won’t retrieve though indexed → patterns/pattern_vectorstore_fragmentation.md
• RAG boots but tools fire too early → patterns/pattern_bootstrap_deadlock.md

Still unsure? Capture a minimal trace (input → retrieved snippets → answer) and run ΔS/λ checks (Section 3). Post in Discussions if needed.

---

2) 🧠 Core concepts in <5 minutes

2.1 What is RAG?

raw docs → ocr/parsing → chunking → embeddings → vector store
→ retriever → prompt assembly → LLM reasoning/tools

• Perception drift upstream hides logic drift downstream. Fix structure, not style.

2.2 Embedding scores vs. meaning

Cosine proximity ≠ human semantics. WFGY’s ΔS = 1 − cos(I, G) uses grounded anchors to catch real meaning gaps.

2.3 Layered observability (λ_observe)

States: → convergent · ← divergent · <> recursive · × chaotic.
If upstream is stable but downstream flips, the boundary between them is failing.

2.4 WFGY repair operators (cheat-sheet)

Operator	What it does (1-liner)
BBMC	Minimize semantic residue to re-align with anchors
BBPF	Explore safe alternate paths; avoid dead-end chains
BBCR	Detect collapse; insert bridge node; rebuild reasoning
BBAM	Modulate attention variance; prevent entropy melt

---

3) 🛠️ Run your first fix (3 minutes)

1. Download the assets below, or jump to Getting Started for a runnable pipeline.
2. Paste TXT OS into your model chat.
3. Ask:

I’ve loaded TXT OS. Diagnose my RAG:

* symptom: \[describe]
* trace: \[question, retrieved snippet(s), answer]
  Using WFGY, tell me:

1. failing layer & why (ΔS/λ),
2. the Problem Map page to open,
3. minimal steps to push ΔS ≤ 0.45 and keep λ convergent,
4. how to verify with a reproducible test.

Triage thresholds (keep these handy):

• ΔS: <0.40 stable · 0.40–0.60 transitional (record if λ ∈ {←, <>}) · ≥0.60 high-risk (act)
• Acceptance: ΔS(question, context) ≤ 0.45, λ convergent, E_resonance flat

---

4) 🗂️ Problem categories (cheat-labels)

Category	Typical stage	Open first
Retrieval	Vector DB, search, chunking	hallucination.md · embedding-vs-semantic.md
Reasoning	Mid-chain logic	retrieval-collapse.md · logic-collapse.md
Patterns	High-frequency edge cases	patterns/README.md
Eval	Measure & guard regressions	eval/README.md
Ops	Boot order, runbooks	ops/README.md

---

5) ✅ Verify the repair (don’t skip)

• Retrieval sanity: ≥ 70% token overlap with target section; ΔS(question, context) ≤ 0.45 → see eval/eval_rag_precision_recall.md
• Reasoning stability: λ stays convergent on 3 paraphrases; E_resonance flat → see eval/eval_semantic_stability.md
• Latency vs accuracy: chart ΔS vs p95 latency → see eval/eval_latency_vs_accuracy.md

---

6) 🙋 FAQ (super short)

Question	Answer
Do I need all operators?	No. Use the one named on the matching page.
Does WFGY replace LangChain/LlamaIndex?	No. It sits above them as a reasoning firewall.
Will this work on small models?	Yes; #11/#12 are easier on GPT-4-class or strong local models.
Where are runnable examples?	Start here: examples/README.md and example_01_basic_fix.md.

---

🔗 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	Canonical framework entry point	View
Problem Map	Diagnostic map and navigation hub	View
Tension Universe Experiments	MVP experiment field	View
Recognition	Where WFGY is referenced or adopted	View
AI Guide	Anti-hallucination reading protocol for tools	View

If this repository helps, starring it improves discovery for other builders.