Skip to content

eval_playbook.md

Latest commit

 

History

History
156 lines (113 loc) · 6.64 KB

eval_playbook.md

File metadata and controls

156 lines (113 loc) · 6.64 KB

Eval Observability — Evaluation Playbook

🧭 Quick Return to Map

You are in a sub-page of Eval_Observability.
To reorient, go back here:

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.

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:

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

Layer Page What it’s for
⭐ Proof WFGY Recognition Map External citations, integrations, and ecosystem proof
⚙️ Engine WFGY 1.0 Original PDF tension engine and early logic sketch (legacy reference)
⚙️ Engine WFGY 2.0 Production tension kernel for RAG and agent systems
⚙️ Engine WFGY 3.0 TXT based Singularity tension engine (131 S class set)
🗺️ Map Problem Map 1.0 Flagship 16 problem RAG failure taxonomy and fix map
🗺️ Map Problem Map 2.0 Global Debug Card for RAG and agent pipeline diagnosis
🗺️ Map Problem Map 3.0 Global AI troubleshooting atlas and failure pattern map
🧰 App TXT OS .txt semantic OS with fast bootstrap
🧰 App Blah Blah Blah Abstract and paradox Q&A built on TXT OS
🧰 App Blur Blur Blur Text to image generation with semantic control
🏡 Onboarding Starter Village Guided entry point for new users

If this repository helped, starring it improves discovery so more builders can find the docs and tools.
GitHub Repo stars