8.5 KiB
Eval Observability — Evaluation Playbook
🧭 Quick Return to Map
You are in a sub-page of Eval_Observability.
To reorient, go back here:
- Eval_Observability — evaluation metrics and system observability
- 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 desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.
Evaluation disclaimer (observability playbook)
The signals in this playbook come from specific logging and probing setups.
They are tools for monitoring behavior, not proofs of safety or correctness on their own.
A compact playbook to stabilize evaluation and ensure results are reproducible.
Use this when metrics look inconsistent, coverage drifts, or benchmarks feel untrustworthy.
Core acceptance
- ΔS(question, retrieved) ≤ 0.45
- Coverage ≥ 0.70 to target section
- λ remains convergent across 3 paraphrases and 2 seeds
- Variance ratio ≤ 0.15 across seed runs
- No monotonic downward drift beyond 3 windows
Step 1 — Regression Gate
Run a gate before promoting new models or retrievers.
- Require ΔS ≤ 0.45 and coverage ≥ 0.70 on 3 paraphrases.
- Block deploy if λ flips divergent.
Open: regression_gate.md
Step 2 — Alerting and Probes
Set live probes to catch collapse early.
- Alert when ΔS ≥ 0.60 or λ flips.
- Emit logs for coverage gaps, drift slopes, entropy rises.
Open: alerting_and_probes.md
Step 3 — Coverage Tracking
Measure retrieved vs target section alignment.
- At least 70% coverage on gold targets.
- Track section-level recall, not just overall hit-rate.
Open: coverage_tracking.md
Step 4 — ΔS Thresholds
Normalize ΔS scoring and enforce thresholds.
- Stable < 0.40
- Transitional 0.40–0.60
- Collapse ≥ 0.60
Open: deltaS_thresholds.md
Step 5 — λ Observe
Use λ probes to detect schema drift.
- Run same query across seeds and paraphrases.
- If λ flips, lock schema and enforce reranker.
Open: lambda_observe.md
Step 6 — Variance and Drift
Log variance ratios and detect slow regressions.
- Variance ratio ≤ 0.15
- Drift slope ≤ 0.02 per batch
- Alert if monotonic decline lasts 3+ windows
Open: variance_and_drift.md
Step 7 — Audit & Escalation
If instability remains:
- Rebuild embeddings with embedding-vs-semantic.md
- Patch retriever fragmentation with vectorstore-fragmentation.md
- Apply reranking schema: rerankers.md
Copy-paste eval contract
eval_contract:
seeds: 3
paraphrases: 3
targets:
deltaS: <=0.45
coverage: >=0.70
lambda: convergent
variance: <=0.15
drift: <=0.02
alerts:
- deltaS >=0.60
- lambda divergent
- drift slope >0.02
🔗 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.
