16 KiB
Retrieval — Global Fix Map
🏥 Quick Return to Emergency Room
You are in a specialist desk.
For full triage and doctors on duty, return here:
- 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 sub-room.
If you want full consultation and prescriptions, go back to the Emergency Room lobby.
Evaluation disclaimer (retrieval)
All retrieval scores and examples in this section come from controlled setups with chosen corpora and prompts.
They help you compare retrieval strategies locally but are not universal rankings of models or vector stores.
A compact hub to stabilize retrieval quality across stacks, models, and stores.
Use this page to route symptoms to the exact structural fix and verify with measurable targets. No infra change required.
Orientation: what each page does
| Page | What it solves | Typical symptom |
|---|---|---|
| Retrieval Playbook | End to end rebuild order and knobs | You fixed one thing and another breaks |
| Retrieval Traceability | Cite-then-explain schema with required fields | Citations miss the exact section or cannot be verified |
| Rerankers | Deterministic reranking across BM25 + ANN | Hybrid worse than single retriever |
| Query Parsing Split | One query, two meanings; detect and route | Answers jump between two unrelated sections |
| Chunk Alignment | Chunking aligned with the model’s semantic window | Snippets cut mid-thought; anchors missing |
| ΔS Probes | Quick health check using ΔS and λ_observe | Looks fine by eye but flips across runs |
| Retrieval Eval Recipes | Deterministic, SDK-free evaluation | No stable way to tell if “better” shipped |
| Store-Agnostic Guardrails | Locks for metrics, analyzers, versions | Index “healthy” but recall still low |
When to use this folder
- High similarity but wrong meaning.
- Correct facts exist in the corpus but never show up.
- Citations inconsistent or missing across steps.
- Hybrid retrieval underperforms a single retriever.
- Index looks healthy while coverage remains low.
Acceptance targets
- ΔS(question, retrieved) ≤ 0.45
- Coverage of target section ≥ 0.70
- λ_observe convergent across 3 paraphrases and 2 seeds
- E_resonance flat on long windows
Symptoms → exact fixes
| Symptom | Likely cause | Open this |
|---|---|---|
| High similarity yet wrong answer | Metric or analyzer mismatch | Embedding ≠ Semantic |
| Correct fact never retrieved | Fragmentation or missing anchors | Vectorstore Fragmentation · Chunking Checklist |
| Hybrid worse than single | Query parsing split or mis-weighted rerank | Query Parsing Split · Rerankers |
| Citations missing or unstable | Schema not enforced | Retrieval Traceability · Data Contracts |
| Answers flip between runs | Prompt header reordering or λ variance | Context Drift · Rerankers |
60-second fix checklist
-
Lock metrics and analyzers
One embedding model per field. One distance metric. Same analyzer for write and read.
Guide: Store-Agnostic Guardrails -
Enforce the snippet contract
Requiresnippet_id,section_id,source_url,offsets,tokens.
Guide: Data Contracts -
Measure ΔS and λ
Run three paraphrases and two seeds.
Guide: ΔS Probes -
Sweep k and rerankers
Try k in {5, 10, 20}. Keep BM25 and ANN candidate lists.
Guide: Rerankers -
Rebuild where needed
Follow the sequence in the playbook and re-test coverage.
Guide: Retrieval Playbook
Checklists — copy before deploy
| Checklist | Scope | Link |
|---|---|---|
| Retrieval Readiness | Pre-flight: embeddings, analyzers, index, gold set | retrieval_readiness.md |
| Reranker Sanity | Hybrid reranking health and overlap checks | reranker_sanity.md |
| Traceability Gate | Contract enforcement for cite-then-explain | traceability_gate.md |
Vector DBs — jump if store specific
-
Family index:
Vector DBs & Stores -
Direct store guides:
FAISS · Chroma · Qdrant · Weaviate · Milvus · pgvector · Redis · Elasticsearch · Pinecone · Typesense · Vespa
Minimal probe pack you can paste
Context: I loaded TXT OS and the WFGY pages.
Task:
- Given question "Q", log ΔS(Q, retrieved) and λ across 3 paraphrases.
- Enforce cite-then-explain with the traceability schema.
- If ΔS ≥ 0.60 or λ flips, return the smallest structural change to push ΔS ≤ 0.45 and coverage ≥ 0.70.
- Use BBMC, BBCR, BBPF, BBAM when relevant.
Return JSON:
{ "citations": [...], "ΔS": 0.xx, "λ_state": "<>", "coverage": 0.xx, "next_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.
