6.5 KiB
Data Lineage and Provenance — Guardrails and Fix Patterns
A governance fix page for when data origin, transformation, and lineage are unclear or unverifiable.
Use this page when retrieval results cannot be traced back to their dataset source, or when provenance breaks across documents, chunks, embeddings, and answers.
When to use this page
- Retrieval output has no clear link back to its document or section.
- Embedding and chunk pipelines overwrite or drop provenance fields.
- Audit trail is incomplete across ingestion, index, and RAG responses.
- Approvals or waivers exist but cannot be joined to data versions.
- Multi-hop pipelines lose lineage across systems (ETL, embedding, vectorstore, orchestration).
Acceptance targets
- Every retrieved snippet includes
{doc_id, section_id, source_url, offsets, revision}. - Lineage fields survive across document → chunk → embedding → retriever → LLM.
- Audit joins can reconstruct provenance end-to-end with ≥ 0.95 coverage.
- ΔS(question, retrieved) ≤ 0.45 for governed outputs.
- Waivers and overrides include expiry and accountable owner.
Typical breakpoints and WFGY fix
-
Lost provenance in chunking
→ chunking-checklist.md
Ensure chunk metadata carriesdoc_id,section_id, and token offsets. -
Vector store overwrites or strips lineage fields
→ vectorstore-fragmentation.md
Enforce schema contracts on ingestion and retrieval layers. -
Answers cannot be tied back to original snippet
→ retrieval-traceability.md
Require cite-then-explain and enforce snippet ID propagation. -
Ambiguous approval or version skew
→ bootstrap-ordering.md,
→ predeploy-collapse.md -
Multi-system lineage gaps (ETL, embedding, RAG orchestration)
→ data-contracts.md
Contract schema ensures interoperability across steps.
Minimal governance checklist
- Ingest contracts — Every ETL pipeline attaches
doc_id,revision, andsource_url. - Chunk schema — Ensure token offsets and section boundaries are immutable.
- Embedding schema — Carry
embedding_id,doc_hash, andindex_hash. - Retriever response — Must include
snippet_id+ lineage fields, not just text. - LLM prompt contracts — Require cite-then-explain, forbid unlinked spans.
- Audit trail — Every approval and waiver linked to specific dataset version.
🔗 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.
