A water utility runs a tabletop that assumes the AI leak-detection model fails. Which exercise outcome best validates BC readiness for that AI failure mode?

A tabletop that assumes AI leak-detection failure is only valuable if it validates that the water utility can sustain detection through non-AI means. Demonstrating failover from the AI model to scheduled manual inspection rounds with clear ownership proves BC readiness: routes, frequencies, accountable supervisors, and criteria for returning to AI-assisted detection after recovery.

Demonstrating failover from the primary AI model to the standby replica model with clear ownership does not validate continuity — a replica shares the same model logic and likely the same failure mode and provides no non-AI way to find leaks. Demonstrating restoration of the AI leak-detection model from versioned backup with clear ownership validates recovery, not continuity; it does not prove the utility can detect leaks while the AI is down and leaves a detection gap until the restore completes. Demonstrating failover from AI work-order prioritization to FIFO dispatch queue with clear ownership keeps dispatch moving but does not replace leak detection, and it abandons real-time leak finding during the outage.

Exam caveat: manual rounds are typically lower coverage or higher latency than a healthy AI detector; the exercise should record residual risk and staffing cost, not pretend parity. Operational check: after the tabletop, publish an after-action item that names the on-call owner for manual inspection activation and the maximum acceptable time from AI failure declaration to first completed manual round.