WFGY/ProblemMap/patterns

Patterns — Failure Catalog (Problem Map 2.0)

This folder is a field guide to recurring failures in RAG and multi-stage LLM pipelines.
Each pattern is actionable: fast signals, root causes, a minimal repro, a deterministic fix, and links to hands-on examples (SDK-free, stdlib-only).

How to use this folder

1. Start with the symptom you’re seeing.
2. Open the matching pattern and run the Minimal Repro + Standard Fix.
3. Wire the acceptance criteria into CI (see Example 08) so the fix stays fixed.

---

Quick Index

Pattern	Problem Map No.	Symptoms you’ll see	Fix entrypoint
RAG Semantic Drift (pattern_rag_semantic_drift.md)	No.1	Plausible but ungrounded answers; citations don’t contain the claim	Example 01, Example 03
Memory Desync (pattern_memory_desync.md)	— (State/Context)	Old names/IDs reappear; agents disagree across turns	Example 04
Vector Store Fragmentation (pattern_vectorstore_fragmentation.md)	No.3	Recall flips across envs; score scales change; rank inversions	Example 05
Hallucination Re-Entry (pattern_hallucination_reentry.md)	— (Provenance)	Model’s prior text shows up as “evidence”; non-corpus sources cited	Example 06
Bootstrap Deadlock (pattern_bootstrap_deadlock.md)	No.14	/readyz stuck/flapping; circular waits at startup	Example 07
Query Parsing Split (pattern_query_parsing_split.md)	— (Parsing)	Multi-intent prompts answered partially or mixed	Example 03, Example 04
Symbolic Constraint Unlock (SCU) (pattern_symbolic_constraint_unlock.md)	No.11 (Symbolic collapse)	“Must/Only/Never” rules vanish mid-pipeline; impossible states	Example 03, Example 04, Example 08

Legend: Problem Map numbers refer to root categories used across the repo. “—” means cross-cutting (not a single number).

---

Pick-a-Pattern in 30 Seconds (Triage Flow)

1. Grounding first — Run Example 01 on a few failing questions.
   • If refusal behavior or citations fail ⇒ go to Semantic Drift.
2. Context/state sanity — Check context_id / mem_rev/hash.
   • Mismatch ⇒ Memory Desync.
3. Index parity — Validate index_out/manifest.json vs runtime.
   • Drift or score scale shift ⇒ Vector Store Fragmentation.
4. Provenance — Inspect source for cited ids.
   • Any model|chat|tmp: ⇒ Hallucination Re-Entry.
5. Startup — If the first minute after deploy is flaky ⇒ Bootstrap Deadlock.
6. Query shape — If the prompt mixes “compare… then draft…” ⇒ Query Parsing Split.
7. Logic rules — If answers cross “must/only/never” boundaries ⇒ SCU.

---

Standard Acceptance Gates (copy to CI)

• Guarded Output: either exact refusal token not in context or JSON with claim + citations:[id,…] scoped to retrieved ids.
• Provenance: all citations pass the corpus-only filter (no chat:/draft:/tmp:).
• Context Consistency: if used, context_id.mem_rev/hash echoes the turn snapshot.
• Constraint Integrity (SCU): constraints_echo ≡ locked set; no contradiction patterns matched.
• Quality Gates (Ex.08): precision≥0.80, under-refusal≤0.05, citation hit rate≥0.75.

---

File Layout

• pattern_rag_semantic_drift.md — How to stop plausible-but-wrong answers with hard grounding.
• pattern_memory_desync.md — One snapshot per turn; bind and echo across agents.
• pattern_vectorstore_fragmentation.md — Keep embeddings/metrics/chunkers aligned.
• pattern_hallucination_reentry.md — Keep model/session text out of evidence.
• pattern_bootstrap_deadlock.md — Deterministic startup ordering and readiness.
• pattern_query_parsing_split.md — Deterministically split multi-intent prompts.
• pattern_symbolic_constraint_unlock.md — Lock+echo constraints; gate contradictions.

See ../examples/ for runnable, stdlib-only code referenced in each pattern.

---

Contributing (tight process)

1. Propose a new pattern via issue labels: pattern-proposal, with minimal repro + acceptance gate.
2. Stabilize with an example (Python or Node, stdlib-only).
3. Add to this README only after approval.
4. Guard with Example 08 metrics before shipping a pattern-driven fix.

---

🔗 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.