Users report wrong answers, but all services look healthy. What do you need?
Why D is the answer
Traces of the agent's steps
Service health does not reveal reasoning or retrieval errors.
In detail
- Service health does not show reasoning or retrieval errors.
- Health checks do not show reasoning.
Use case
An illustrative scenario, not a real incident.
All services are healthy, but answers are wrong. Traces show a bad retrieval step.
See it in different ways
flowchart LR A["Agent run"] --> B["Trace of each step"] B --> C["Dashboards and alerts"] C --> D["Root cause"] D --> E["Add a test and fix"]
stateDiagram-v2 [*] --> Running Running --> Incident Incident --> ReadTrace ReadTrace --> Fix Fix --> AddEvalCase AddEvalCase --> StagedRollout StagedRollout --> Running
flowchart TD
Q{"What do you need?"}
Q -->|"Service metrics"| A0["Show only uptime"]
Q -->|"Agent traces"| A1["Show each step"]Key point: Traces show what happened.
Why the other options are wrong
- A. A CPU temperature graph
- CPU shows only load.
- B. A larger disk on the load balancer
- Disk shows only capacity.
- C. A second CPU metrics dashboard
- CPU metrics show only load.