mirror of
https://github.com/onestardao/WFGY.git
synced 2026-05-04 22:41:17 +00:00
134 lines
5.4 KiB
Markdown
134 lines
5.4 KiB
Markdown
# Eval Observability — Alerting and Probes
|
||
|
||
<details>
|
||
<summary><strong>🧭 Quick Return to Map</strong></summary>
|
||
|
||
<br>
|
||
|
||
> You are in a sub-page of **Eval_Observability**.
|
||
> To reorient, go back here:
|
||
>
|
||
> - [**Eval_Observability** — evaluation metrics and system observability](./README.md)
|
||
> - [**WFGY Global Fix Map** — main Emergency Room, 300+ structured fixes](../README.md)
|
||
> - [**WFGY Problem Map 1.0** — 16 reproducible failure modes](../../README.md)
|
||
>
|
||
> 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.
|
||
</details>
|
||
|
||
|
||
A live guardrail system that detects semantic drift, retrieval collapse, and logic instability in production.
|
||
Use this page to design **continuous probes** (ΔS, λ, coverage, resonance) and trigger alerts before users see failures.
|
||
|
||
---
|
||
|
||
## Why probes are required
|
||
|
||
- **Silent regressions**: Models may degrade gradually after retraining or infra changes.
|
||
- **Runtime entropy**: Long chains often destabilize after 25–40 steps.
|
||
- **Hybrid stack drift**: Store upgrades, reranker weights, or tokenizer shifts silently change outcomes.
|
||
- **Auditability**: Real-time probes make every failure reproducible.
|
||
|
||
---
|
||
|
||
## Core probe dimensions
|
||
|
||
| Metric | Probe type | Alert condition |
|
||
|--------|------------|-----------------|
|
||
| ΔS(question, retrieved) | per-query | ≥ 0.60 for any query OR average >0.50 across batch |
|
||
| Coverage of target section | per-query | < 0.70 for more than 5% of batch |
|
||
| λ_observe | rolling window | divergence observed across 2 seeds or paraphrases |
|
||
| E_resonance | sliding horizon | spikes or oscillations in 50–100 step runs |
|
||
|
||
---
|
||
|
||
## Recommended probe architecture
|
||
|
||
1. **Pre-query probe**
|
||
Check retriever config hash, index hash, analyzer type.
|
||
Block if mismatched against gold baseline.
|
||
|
||
2. **Mid-query probe**
|
||
During retrieval, compute ΔS(question, retrieved).
|
||
Attach snippet IDs and offsets for traceability.
|
||
|
||
3. **Post-query probe**
|
||
Run λ stability check across 3 paraphrases.
|
||
If divergent, tag output with `unstable=true`.
|
||
|
||
4. **Long-chain probe**
|
||
For conversations >25 steps, sample entropy and E_resonance every 10 steps.
|
||
Trigger backoff or memory split if instability rises.
|
||
|
||
---
|
||
|
||
## Alerting routes
|
||
|
||
- **Slack / Teams** → Send structured JSON logs with ΔS, λ, coverage, index hash, retriever config.
|
||
- **PagerDuty** → Trigger incidents only when threshold failures exceed N in M minutes.
|
||
- **Dashboards** → Grafana/Datadog with ΔS trend lines, λ variance plots, coverage histograms.
|
||
- **Audit store** → Write all probe outputs to KV or DB keyed by `(session_id, index_hash, retriever_config)`.
|
||
|
||
---
|
||
|
||
## Example probe config (YAML)
|
||
|
||
```yaml
|
||
probes:
|
||
deltaS:
|
||
threshold: 0.60
|
||
action: alert
|
||
coverage:
|
||
threshold: 0.70
|
||
tolerance: 0.05
|
||
lambda:
|
||
seeds: 2
|
||
paraphrases: 3
|
||
action: block
|
||
resonance:
|
||
horizon: 100
|
||
action: degrade_notice
|
||
alerting:
|
||
sink:
|
||
- slack
|
||
- pagerduty
|
||
- grafana
|
||
````
|
||
|
||
---
|
||
|
||
## Common mistakes
|
||
|
||
* **Only probing ΔS**. Always include coverage + λ + resonance.
|
||
* **Static thresholds**. Thresholds must be tested against gold sets and updated quarterly.
|
||
* **No audit linkage**. Alerts without index hash and retriever config cannot be replayed.
|
||
* **Flooding alerts**. Use capped retries and aggregate rules to avoid pager fatigue.
|
||
|
||
---
|
||
|
||
### 🔗 Quick-Start Downloads (60 sec)
|
||
|
||
| Tool | Link | 3-Step Setup |
|
||
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------- |
|
||
| **WFGY 1.0 PDF** | [Engine Paper](https://github.com/onestardao/WFGY/blob/main/I_am_not_lizardman/WFGY_All_Principles_Return_to_One_v1.0_PSBigBig_Public.pdf) | 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + \<your question>” |
|
||
| **TXT OS (plain-text OS)** | [TXTOS.txt](https://github.com/onestardao/WFGY/blob/main/OS/TXTOS.txt) | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
|
||
|
||
---
|
||
|
||
<!-- WFGY_FOOTER_START -->
|
||
|
||
### Explore More
|
||
|
||
| Module | Description | Link |
|
||
| --- | --- | --- |
|
||
| WFGY Core | Canonical framework entry point | [View](https://github.com/onestardao/WFGY/tree/main/core/README.md) |
|
||
| Problem Map | Diagnostic map and navigation hub | [View](https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md) |
|
||
| Tension Universe Experiments | MVP experiment field | [View](https://github.com/onestardao/WFGY/tree/main/TensionUniverse/Experiments) |
|
||
| Recognition | Where WFGY is referenced or adopted | [View](https://github.com/onestardao/WFGY/blob/main/recognition/README.md) |
|
||
| AI Guide | Anti-hallucination reading protocol for tools | [View](https://github.com/onestardao/WFGY/blob/main/AI_GUIDE.md) |
|
||
|
||
> If this repository helps, starring it improves discovery for other builders.
|
||
> [](https://github.com/onestardao/WFGY)
|
||
|
||
<!-- WFGY_FOOTER_END -->
|
||
|