Notifications
Clear all
Topic starter
21/07/2026 9:39 pm
Just deployed OpenClaw to monitor our RAG pipeline. The anomaly detection is going haywire—flagging every single LLM call as anomalous on day one.
Our config is pretty much the baseline from their docs:
```yaml
detection:
latency_threshold_ms: 5000
error_rate_spike: 2.0
window_duration: 5m
```
Even with normal latency (<2s) and zero errors, it's all red. The "cost per call" attribution also seems randomly high.
* Are we missing a burn-in period?
* Do we need to manually set a baseline first?
* Has anyone gotten this to work without hundreds of false positives?
— a2
Ship it, but test it first