Edge Decision Aggregation Index (EDAI)

GMI-AR-022Asset ReliabilityInfrastructure-Dependent

Definition

Edge Decision Aggregation Index (EDAI) measures whether the collective effect of edge-level autonomous decisions, such as threshold adjustments, alarm suppressions, and monitoring frequency changes, aligns with the centrally defined reliability program standards.

The Measurement Gap

A utility managing 4,200 distribution transformers had deployed edge analytics on 800 of them: smart sensors with embedded processing that monitored dissolved gas analysis, load profiles, and temperature. Each edge device had limited autonomous authority. It could adjust its own monitoring frequency based on detected conditions, suppress low-severity alarms that met certain patterns, and modify alert thresholds within a defined band.

These were sensible capabilities. A transformer experiencing seasonal load changes doesn't need the same alert threshold in January as in July. An edge device that recognizes a known-benign transient pattern shouldn't wake up the operations center at 3 AM. The edge decisions were locally optimal.

After 18 months, the fleet reliability engineer noticed something she couldn't explain from the central dashboard. Transformer failure rates in the edge-monitored fleet were slightly higher than in the non-monitored fleet, the opposite of what a monitoring investment should produce. Not dramatically higher, but 8% higher than the unmonitored baseline.

She dug into the edge logs. What she found was instructive. Across the 800 edge devices, alarm thresholds had drifted. Each device had individually adjusted its thresholds in response to local conditions. Some had raised thresholds to suppress seasonal false alarms, others had lowered them in response to transient events. In aggregate, the average alarm threshold across the fleet had risen 23% above the centrally defined baseline. The edge devices had collectively desensitized themselves.

Additionally, alarm suppression patterns had evolved. Edge devices were suppressing an average of 34% of alarms, alarms that would have been routed to the operations center under the original configuration. Most of those suppressions were appropriate, matching known-benign patterns. But approximately 11% were genuine precursor events that the edge suppression logic had incorrectly classified as benign.

No single edge device's behavior was alarming. Each had made individually reasonable adjustments. But in aggregate, the fleet had quietly raised its detection threshold by 23% and was suppressing 11% of genuine alerts. The central monitoring team had no visibility into this aggregate drift because they only saw the alerts that survived edge filtering. They didn't know what they weren't seeing.

Asset Availability showed the fleet averages. Alarm rate actually looked better, because fewer alarms were reaching the control center, which everyone interpreted as a sign the fleet was healthy. Transformer failure rate captured the elevated failures but couldn't trace them to edge-level autonomous decisions that had collectively degraded the monitoring program.

There is no metric that tracks the cumulative impact of edge-level autonomous decisions on the overall reliability program.

Formula

Edge Decision Aggregation Index (EDAI) measures whether the collective effect of edge-level autonomous decisions aligns with centrally defined reliability program standards.

In plain language:

Are all our edge devices, taken together, still doing what the reliability program intended, or have they collectively drifted somewhere else?

The Formula

EDAI = 1 - (1/N) x Sum_n |E_n - G_n| / G_n

Where N is the number of edge decision categories tracked, E_n is the aggregate edge-enacted value for category n, and G_n is the global program-intended value for category n.

Worked Example

The transformer utility tracks three edge decision categories:

| Category | Program Standard (G_n) | Fleet Avg Edge Value (E_n) | Deviation |
| DGA alarm threshold (ppm) | 150 | 184.5 | 23.0% |
| Alarm suppression rate | 15% | 34.2% | 128.0% |
| Monitoring frequency (readings/day) | 24 | 18.7 | 22.1% |

EDAI = 1 - (0.230 + 1.280 + 0.221) / 3 = 1 - 0.577 = 0.423

An EDAI of 0.423 is critically low. The edge fleet has massively diverged from the program's intent. The alarm suppression rate is the primary offender: edge devices are suppressing more than double the intended rate. Alarm thresholds and monitoring frequency have both drifted more than 20% from their design points.

This EDAI makes the invisible problem visible. The individual edge devices each made locally reasonable decisions. In aggregate, they've degraded the monitoring program's effectiveness by widening detection gaps and reducing data collection density.

Sign Convention

Higher is better. 0.85 and above is well-aligned. 0.70 to 0.85 needs review. Below 0.70 indicates significant program-to-edge misalignment. These thresholds are proposed, not proven.

Cross-References

Upstream: Autonomy Boundary Calibration Score, because autonomy boundaries govern what edges are allowed to decide. If that score shows overly permissive boundaries, EDAI will reflect the drift those boundaries allow. Health Score Contextual Accuracy, because if health scores at the edge are contextually inaccurate, edge decisions based on those scores will be misguided.

Downstream: Autonomous WO Signal-to-Noise Ratio, because edge-adjusted thresholds and suppression rates directly affect which events become work orders and which are silently discarded. Operator Trust Calibration Score, because edge-suppressed alarms that should have reached operators affect trust dynamics and response quality.

EDAI is the fleet-level companion to the Autonomy Boundary Calibration Score's individual-system focus. That score asks whether this system's authority is correctly set. EDAI asks what happens when 800 systems each exercise their authority independently.

Proposed formula, pending validation pilots.

At a Glance

Category
System Feedback and Fleet Integrity

Decision Supported

Are my edge devices collectively making decisions that align with the reliability program, or are locally optimal edge decisions creating globally suboptimal outcomes that no one is tracking?

← Cohort Homogeneity Score (CHS)Health Score Contextual Accuracy (HSCA) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment