fix(ai): filter out noise from anomalies and status changes

Critical changes to surface only actionable insights:

1. Anomalies now require at least 50% deviation from baseline
   - '1.0x baseline' values filtered out (statistically significant but not actionable)
   - Must be >1.5x above OR <0.5x below baseline to report

2. Status changes filter out startup noise
   - 'unknown → running' is just system starting, not a real state change
   - Backups removed from main list (they have dedicated section)

3. Only show genuinely interesting changes:
   - Config changes, migrations, restarts, deletions
   - Things that require operator attention

This massively reduces noise while keeping high-signal alerts.

frontend-modern/src/components/shared/AIOverviewTable.tsx

@@ -291,14 +291,22 @@ export const AIOverviewTable: Component<{ showWhenEmpty?: boolean }> = (props) =
         }
 
 
-        // Changes - only show meaningful changes, not just "created" (infrastructure detection)
+        // Changes - only show meaningful changes, not noise
-        // Valuable changes: config, status, migrated, restarted, deleted, backed_up
+        // Skip: created (just patrol discovering resources)
-        // Skip: created (just means AI patrol discovered the resource)
+        // Skip: backed_up (backups have their own section, no need to duplicate here)
-        const meaningfulChanges = changes().filter(c => c.change_type !== 'created');
+        // Skip: status changes from "unknown" (just startup, not real state changes)
+        const meaningfulChanges = changes().filter(c => {
+            if (c.change_type === 'created') return false;
+            if (c.change_type === 'backed_up') return false;
+            // Filter out startup noise: "unknown → running" or "unknown → stopped"
+            if (c.change_type === 'status' && c.description?.includes('unknown →')) return false;
+            return true;
+        });
         const visibleChanges = showAllChanges()
             ? meaningfulChanges
             : meaningfulChanges.slice(0, CHANGE_LIMIT);
 
+
         for (const change of visibleChanges) {
             const changeTypeBadgeClass: Record<string, string> = {
                 deleted: 'bg-red-100 text-red-700 dark:bg-red-900/40 dark:text-red-300',

internal/ai/baseline/store.go

@@ -291,33 +291,40 @@ func (s *Store) CheckResourceAnomalies(resourceID string, metrics map[string]flo
 	
 	for metric, value := range metrics {
 		severity, zScore, baseline := s.CheckAnomaly(resourceID, metric, value)
-		if severity != AnomalyNone {
+		if severity != AnomalyNone && baseline != nil {
+			// Compute ratio: current value / baseline mean
+			ratio := value / baseline.Mean
+			
+			// Filter out statistically significant but practically meaningless anomalies
+			// Users don't care about "1.0x baseline" or small deviations
+			// Require at least 50% deviation from baseline to report
+			if ratio >= 0.5 && ratio <= 1.5 {
+				continue // Too close to baseline to be actionable
+			}
+			
 			report := AnomalyReport{
 				ResourceID:   resourceID,
 				Metric:       metric,
 				CurrentValue: value,
 				ZScore:       zScore,
 				Severity:     severity,
+				BaselineMean: baseline.Mean,
+				BaselineStdDev: baseline.StdDev,
 			}
 			
-			if baseline != nil {
+			// Generate human-readable description
-				report.BaselineMean = baseline.Mean
+			direction := "above"
-				report.BaselineStdDev = baseline.StdDev
+			if zScore < 0 {
-				
+				direction = "below"
-				// Generate human-readable description
-				ratio := value / baseline.Mean
-				direction := "above"
-				if zScore < 0 {
-					direction = "below"
-				}
-				report.Description = formatAnomalyDescription(metric, ratio, direction, severity)
 			}
+			report.Description = formatAnomalyDescription(metric, ratio, direction, severity)
 			
 			anomalies = append(anomalies, report)
 		}
 	}
 	
 	return anomalies
+
 }
 
 // formatAnomalyDescription generates a human-readable anomaly description