Definition
Percentage of AI-generated clearance boundaries that correctly account for all active temporary modifications, cross-tie configurations, and in-progress design changes affecting the isolation logic at the time of clearance issuance.
The Measurement Gap
Equipment isolation (clearance) boundaries in nuclear plants must account for the current plant configuration, including all active temporary modifications, cross-tie configurations, and in-progress design changes. When AI systems generate clearance boundaries, they must access current configuration data to produce boundaries that reflect the actual plant state. No legacy metric measures whether AI-generated clearance boundaries correctly account for the current modification status. Clearance program metrics track clearance processing time, hanging/clearing efficiency, and clearance-related events. Human performance error rate captures execution errors. None of them measure whether the AI's boundary recommendation reflects the actual plant configuration at the time the clearance was issued. The invisible failure mode is that the AI generates a boundary based on the design-basis configuration while one or more temporary modifications, cross-ties, or design changes have altered the isolation logic, creating a boundary that looks correct on the drawing but does not isolate the equipment in the as-modified plant.
Decision Supported
Can the plant trust AI-generated clearance boundaries for equipment isolation, or does every AI-recommended boundary require independent walkdown verification against the current modification status?
Formula
Plain language: For each AI-generated clearance boundary, verify whether the boundary accounts for every active temporary modification, installed cross-tie, and in-progress design change that affects the isolation points or flow paths in the boundary. Score each boundary as: Current (all configuration changes accounted for), Partially Current (some changes accounted for, others missed but no safety-significant flow path affected), or Non-Current (configuration change missed that creates or could create an unintended flow path through the isolation boundary).
Notation:
CBCCR = (Current + 0.5 × Partially_Current) / N_boundaries
Variables:
- Current = AI clearance boundaries that correctly account for all active configuration changes (count)
- Partially_Current = boundaries that missed a configuration change but the missed change doesn't affect a safety-significant flow path (count)
- Non_Current = boundaries that missed a configuration change affecting a safety-significant flow path (count)
- N_boundaries = total AI-generated clearance boundaries evaluated in the period (count)
- The 0.5 weight on Partially_Current reflects that even non-safety-significant misses indicate the AI's configuration awareness is incomplete
Sign convention: CBCCR = 1.0 means every boundary was fully current. Lower is worse. Any Non_Current boundary is a potential personnel safety or plant event precursor.
Aggregation period: Weekly rolling (clearances are issued continuously; problems must be detected fast).
Edge Cases
- Temporary modification installed after the AI generated the boundary but before the clearance was issued: track as a "timing gap". The AI was correct at generation time but the configuration changed before issuance. This is a process timing issue, not an AI accuracy issue, but it's still a safety concern. Track separately.
- Clearance for a system with no active temp mods or cross-ties: these are trivially current. Include in the denominator (they establish the baseline accuracy for simple boundaries) but don't weight them in trend analysis
- Minimum sample: all AI-generated clearances (this is safety-critical. No sampling, census measurement)
Proposed formula, pending validation pilots.