Definition
Coefficient of variation of AI-recommended intervals for the same PMS task across consecutive adjustments.
The Measurement Gap
What This Metric Measures
PISI measures the coefficient of variation of AI-recommended intervals for the same PMS task across consecutive adjustment cycles. It quantifies whether the adaptive system produces stable, plannable intervals or oscillating, unpredictable ones that prevent the chief engineer from reliably planning crew workload, spare parts staging, and port selection for maintenance execution.
The Decision It Supports
Is the adaptive PMS system producing intervals that the chief engineer can plan around, or is it changing its mind so frequently that planning becomes impossible? A system that recommends 6,000 hours, then 4,500, then 7,200, then 5,100 for the same task across four adjustment cycles is condition-responsive on paper but operationally useless. The chief engineer cannot pre-position parts for a maintenance window that keeps moving. PISI separates genuinely adaptive scheduling from indecisive oscillation.
Why Legacy Metrics Cannot See This
PMS metrics track the current interval and whether the task is completed on time. They do not track whether the interval itself is stable across adjustment cycles. Each individual adjustment may be condition-justified at the moment it is made. But the pattern of adjustments (extending, shortening, extending, shortening) creates planning instability that no single-point metric can detect. PISI applies a statistical lens to the adjustment history, revealing whether the adaptive system produces reliable schedules or volatile ones.
Formula
Primary FormulaPISI = sigma(intervals) / mu(intervals)
Variable Definitions
- intervals = set of AI-recommended intervals for a specific PMS task across the last N adjustment cycles- sigma = standard deviation of the interval values- mu = mean of the interval values
- N = minimum 4 adjustment cycles for statistical meaning
Units: Dimensionless coefficient of variation. A value of 0 means perfect stability where every adjustment produces the same interval. Higher values indicate more volatility. Compute per PMS task and aggregate to a fleet-level distribution showing the percentage of tasks at each stability level.
Edge Cases
Legitimate condition changes should cause interval changes: PISI does not distinguish between justified and unjustified volatility on its own. Use alongside AIOVR (GMI-MAR-022) to separate oscillation-because-the-data-is-noisy from oscillation-because-the-equipment-condition-is-genuinely-changing. Tasks adjusted fewer than 4 times have insufficient data and should be excluded from PISI computation. Minimum sample of 30 PMS tasks with 4 or more adjustment cycles is required for a meaningful fleet-level distribution.
Proposed formula, pending validation pilots.