Definition
SAAA measures the percentage of SCADA anomaly alerts correctly attributed to their actual source category: cyber intrusion, legitimate-but-unusual operation, sensor/communication fault, or software bug.
The Measurement Gap
Measurement Gap and Decision Supported
The Invisible Failure Mode: A water utility deploys an OT security monitoring platform that detects SCADA anomalies and attributes them to categories: cyber, operational, sensor, or software. Over 12 months, the system generates 1,847 alerts. The security team investigates 312 of these and dismisses the remainder as low-priority. Among investigated alerts, the system's overall attribution accuracy is 78%. But the per-category accuracy reveals a critical gap: the system correctly identifies 91% of sensor faults and 84% of operational anomalies but only 62% of actual cyber events. The 38% cyber misattribution rate means that 4 out of 10 genuine cyber anomalies are classified as something else (typically operational) and receive operational rather than security investigation. The cyber events are hiding in plain sight, labeled as routine.
What Existing Metrics Miss:
OT security platforms report detection rates and alert volumes but rarely report attribution accuracy by category. A system that detects 95% of anomalies but correctly attributes only 62% of cyber events provides a false sense of security. Without per-category SAAA measurement, the most dangerous attribution failures (cyber events classified as non-threats) are invisible in aggregate accuracy metrics.
The Decision SAAA Enables:
When the OT security system attributes an anomaly to a non-cyber category, can the security team trust that attribution, or does the cyber misattribution rate require secondary review of all anomalies regardless of initial attribution?
Cross-References:
SAAA connects downstream to incident response effectiveness. Correct attribution is the prerequisite for correct response selection. SAAA connects to Cybersecurity Analyst Response Fatigue Index (CSARF, GMI-WT-033) because high false positive rates across all categories drive analyst fatigue, which in turn degrades the quality of investigation that feeds SAAA validation. Baseline Integrity Confidence (BIC, GMI-WT-032) is foundational to SAAA. If the anomaly detection model's training baseline was compromised, all attributions derived from that baseline are unreliable.
Formula
Plain language: Sample anomaly alerts from the SCADA security AI. For each, determine the actual root cause through investigation. Compare the AI's attribution to the actual source. Compute accuracy per category.
Notation:
SAAA = Correctly_attributed / Total_investigated
Per-category:
SAAA_cyber = Correctly_attributed_cyber / Total_actual_cyber
SAAA_operational = Correctly_attributed_operational / Total_actual_operational
SAAA_sensor = Correctly_attributed_sensor / Total_actual_sensor
Where:
- Correctly_attributed = alerts where the AI's source category matched the investigation finding
- Total_investigated = alerts that received investigation follow-up with a determined root cause
Units: Percentage overall and per category. The overall rate matters less than the per-category rates. Misattributing cyber events as operational (missing a real threat) is far worse than the reverse.
Proposed formula, pending validation pilots.