# Deployment and Monitoring Checklist

## Release readiness
- [ ] The approved model, prompt, data and code versions are recorded.
- [ ] Automated tests and evaluation thresholds pass.
- [ ] The deployment environment matches the tested environment.
- [ ] Secrets, authentication and permissions are protected.
- [ ] A rollback or safe shutdown path is tested.

## Serving and performance
- [ ] Input validation and output schema checks exist.
- [ ] Latency, throughput, availability and cost targets are defined.
- [ ] Timeouts, retries and error handling are tested.
- [ ] Logs avoid unnecessary personal or confidential data.

## Model and data monitoring
- [ ] Input quality and Data Drift signals are monitored.
- [ ] Prediction distribution and failure rates are monitored.
- [ ] Ground truth or human feedback is collected where possible.
- [ ] Performance is reviewed across important groups and conditions.

## Generative Ai controls
- [ ] Prompt versions and model routes are recorded.
- [ ] Retrieval quality and groundedness are evaluated.
- [ ] Hallucination and unsafe-output signals are monitored.
- [ ] Tool actions use restricted permissions and approval gates.

## Ownership and response
- [ ] Alert thresholds and responsible people are named.
- [ ] Incident, escalation and user-communication procedures exist.
- [ ] Retraining or update triggers are evidence-based.
- [ ] Scheduled review date is recorded.