7.5 KiB
Risk Register and Waivers — Guardrails and Fix Patterns
This page ensures all known risks and temporary waivers in AI pipelines are documented, owned, and auditable.
Use this when risks are acknowledged but not tracked, lack expiry, or are invisible to audits and downstream fixes.
When to use this page
- Waivers are granted informally without expiry.
- Known risks are not logged in a central register.
- Incident or eval failures repeat due to ignored waivers.
- Risk acceptance lacks linkage to Problem Map fixes.
- Ownership for waivers or risks is missing.
Acceptance targets
- 100% of waivers have an owner, rationale, and expiry date.
- All risks logged in a register that maps back to Problem Map page.
- Coverage ≥ 0.95: every model, dataset, prompt, eval has risk review.
- ΔS(question, retrieved) ≤ 0.45 for governed outputs with waivers.
- Immutable log of all waivers linked to audit trail.
Typical breakpoints and WFGY fix
-
Waivers without expiry or owner
→ policy_baseline.md
Require baseline policy: every waiver must expire and have an owner. -
Risks tracked but disconnected from fixes
→ rag-architecture-and-recovery.md
Require mapping each risk to a Problem Map ID. -
Duplicate risks or missing register entries
→ audit_logs_and_traceability.md
Enforce immutable logs cross-checked with lineage.
Minimal risk register schema
| Field | Description |
|---|---|
| Risk ID | Unique identifier (RR-YYYYMMDD-###) |
| Description | Concise, technical summary of risk |
| Linked Problem Map Fix | Which Problem Map No.X mitigates this |
| Owner | Individual accountable for mitigation or acceptance |
| Expiry Date | Mandatory review/renewal date |
| Status | Open / Mitigated / Expired / Waived |
| Evidence Link | Eval result, lineage log, or waiver doc |
Example waiver entry
**Risk ID**: RR-2025-08-001
**Description**: Retrieval drift due to hybrid reranker instability.
**Linked Problem Map Fix**: No.7 (Reranker drift) → [rerankers.md](https://github.com/onestardao/WFGY/blob/main/ProblemMap/rerankers.md)
**Owner**: ML Ops lead (Alice K.)
**Expiry Date**: 2025-09-30
**Status**: Waived (short-term)
**Evidence**: Incident 2025-08-22, eval ΔS=0.58, coverage=0.66. Logged waiver until new reranker release.
Escalation policy
- Expired waivers auto-escalate to governance board.
- Unowned risks block model release.
- Mitigated risks require validation: ΔS ≤ 0.45, λ convergent across 3 probes.
- Waiver renewals must include evidence of attempted fix or mitigation.
🔗 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.
