# Answer Health Monitoring Checklist

## System Health

- Latency, errors, throughput, dependency failures, and token use are tracked.
- Cost is visible by application, environment, team, or use case.
- Alerts distinguish service failure from answer-quality concerns.

## Answer Health

- Groundedness, relevance, completeness, safety, and task success are measured.
- Negative feedback, corrections, regenerations, and escalations are reviewed.
- Evaluation regressions are tracked after prompt, model, retrieval, or data changes.

## Improvement Loop

- Product, engineering, and risk owners review quality signals on a defined cadence.
- New failure examples are added to the evaluation set.
- High-risk outcomes have a human review or escalation path.