# Model operations runbook

Use this runbook to operate model-backed services as managed production systems.

## Daily checks

- Provider availability and fallback path health.
- Cost per task and token budget drift.
- Latency by route, model, tool, and retrieval step.
- Evaluation failures after prompt, model, or tool changes.
- Incident queue, degraded-mode events, and user-visible errors.

## Release checks

Every model or prompt change needs a reason, test result, rollback path, and owner approval.
