Standard operating sequence

1. Readiness check: /health/ready
2. Start run with explicit goal and mode
3. Monitor /api/v1/events and /api/v1/runs/{id}/graph
4. Apply pause/resume/cancel controls as needed
5. Archive evidence (events, graph, attempts)

Incident handling

• Use pause first when uncertainty is high.
• Cancel when policy breach or uncontrolled retry emerges.
• Attach evidence paths to postmortem timeline.

Gate-minded operations

• Quality target >= 0.90
• Time improvement >= 25%
• Cost improvement >= 20%
• Regression must remain false