Definition
Percentage of AI troubleshooting recommendations that account for the vessel complete recent maintenance history.
The Measurement Gap
What This Metric Measures
AI Troubleshooting Recommendation Completeness measures the proportion of AI-generated troubleshooting recommendations that provide genuinely new diagnostic value rather than duplicating work the vessel crew has already performed within the recent history window. The metric captures whether the AI system has sufficient awareness of recent maintenance actions (from structured PMS completions, chief engineer reports, and port service records) to avoid recommending checks, replacements, or inspections that were just done.
The Decision It Supports
This metric drives a critical trust decision: should the chief engineer follow AI troubleshooting guidance or treat it as unreliable? When 1 in 5 recommendations tells the crew to do something they did last week, the entire AI advisory system loses credibility. Operations managers use this metric to decide whether to invest in better data integration for the AI system or to downgrade AI recommendations from actionable guidance to optional reference material.
Why Legacy Metrics Cannot See This
Traditional troubleshooting effectiveness metrics (if they exist at all in maritime operations) measure whether problems get resolved, not whether the diagnostic path was efficient. Legacy systems have no concept of recommendation freshness because human troubleshooting guides are static documents that do not adapt to recent maintenance history. When a technical manual says "check the fuel filter," it says that regardless of whether the filter was replaced yesterday. AI troubleshooting systems promise context-aware recommendations, but without measuring completeness against recent work history, fleets cannot distinguish between an AI system that truly understands vessel maintenance state and one that is essentially generating static checklists with a modern interface. This metric makes that distinction measurable.
Formula
Primary Formula
ATRC = 1 − (Recommendations_Duplicating_Recent_Work / Total_Recommendations)
Variable Definitions
Recommendations_Duplicating_Recent_Work: Number of AI-generated troubleshooting recommendations that suggest actions already completed within the recent history window (N days). Unit: count.
Total_Recommendations: Total number of AI-generated troubleshooting recommendations issued during the measurement period. Unit: count.
N (Recent History Window): Rolling lookback period for determining whether work was recently performed. Default: 30 days.
Units:
Ratio expressed as percentage (0–100%). Higher values indicate better recommendation completeness: 95% means only 5% of recommendations duplicate recent work.
Edge Cases
A re-check recommendation may be legitimately correct if recent work itself introduced a new problem: for example, recommending a fuel pressure check after a recent fuel injector replacement that may have disturbed fuel system integrity. These should be flagged but not automatically counted as duplicates. Partial duplicates (where the AI recommends a broader inspection that includes a recently completed subtask) require judgment on whether the non-duplicate portion justifies the recommendation. Require minimum 20 troubleshooting events per vessel per quarter for statistical validity. The 30-day default window should be adjusted for equipment with longer maintenance intervals.
Proposed formula, pending validation pilots.