diff --git a/backend/rating.go b/backend/rating.go
index c589ffb..5144447 100644
--- a/backend/rating.go
+++ b/backend/rating.go
@@ -203,15 +203,15 @@ func clothingSeverityWeight(label string) float64 {
 
 	switch label {
 	case "lingerie":
-		return 0.64
+		return 0.50
 	case "panties", "bra":
-		return 0.58
+		return 0.44
 	case "bikini":
-		return 0.42
+		return 0.30
 	case "stockings", "heels":
-		return 0.40
+		return 0.28
 	case "skirt", "dress", "hotpants", "croptop":
-		return 0.34
+		return 0.22
 	default:
 		return 0.00
 	}
@@ -431,20 +431,21 @@ func contextualSegmentSeverityWeight(label string) float64 {
 	var score float64
 
 	if set.HasPosition {
+		// Priorität: Position > Kleidung > Objekte (Kombination erhöht den Score).
 		score =
 			0.66*set.Position +
-				0.17*set.Body +
-				0.10*set.Object +
-				0.05*set.Clothing +
+				0.15*set.Body +
+				0.12*set.Clothing +
+				0.05*set.Object +
 				0.02*set.Person
 
 		if set.HasBody {
 			score += 0.055
 		}
-		if set.HasObject {
-			score += 0.040
-		}
 		if set.HasClothing {
+			score += 0.030
+		}
+		if set.HasObject {
 			score += 0.020
 		}
 		if set.HasPerson {
@@ -461,29 +462,27 @@ func contextualSegmentSeverityWeight(label string) float64 {
 
 	score =
 		0.54*set.Body +
-			0.30*set.Object +
-			0.14*set.Clothing +
+			0.28*set.Clothing +
+			0.16*set.Object +
 			0.02*set.Person
 
-	if set.HasBody && set.HasObject {
-		score += 0.09
-	}
 	if set.HasBody && set.HasClothing {
+		score += 0.05
+	}
+	if set.HasBody && set.HasObject {
 		score += 0.04
 	}
-	if set.HasObject && set.HasClothing {
+	if set.HasClothing && set.HasObject {
 		score += 0.03
 	}
 	if set.HasPerson && (set.HasBody || set.HasObject || set.HasClothing) {
 		score += 0.015
 	}
 
-	// Kleidung alleine soll interessant sein können, aber nicht stark.
 	if set.HasClothing && !set.HasBody && !set.HasObject {
 		score = math.Min(score, 0.38)
 	}
 
-	// Ohne Position nicht zu aggressiv.
 	score = math.Min(score, 0.72)
 
 	return ratingClamp01(score)
@@ -524,18 +523,18 @@ func contextualSegmentSeverityWeightFromSet(set ratingSignalSet) float64 {
 	if set.HasPosition {
 		score =
 			0.66*set.Position +
-				0.17*set.Body +
-				0.10*set.Object +
-				0.05*set.Clothing +
+				0.15*set.Body +
+				0.12*set.Clothing +
+				0.05*set.Object +
 				0.02*set.Person
 
 		if set.HasBody {
 			score += 0.055
 		}
-		if set.HasObject {
-			score += 0.040
-		}
 		if set.HasClothing {
+			score += 0.030
+		}
+		if set.HasObject {
 			score += 0.020
 		}
 		if set.HasPerson {
@@ -551,17 +550,17 @@ func contextualSegmentSeverityWeightFromSet(set ratingSignalSet) float64 {
 
 	score =
 		0.54*set.Body +
-			0.30*set.Object +
-			0.14*set.Clothing +
+			0.28*set.Clothing +
+			0.16*set.Object +
 			0.02*set.Person
 
-	if set.HasBody && set.HasObject {
-		score += 0.09
-	}
 	if set.HasBody && set.HasClothing {
+		score += 0.05
+	}
+	if set.HasBody && set.HasObject {
 		score += 0.04
 	}
-	if set.HasObject && set.HasClothing {
+	if set.HasClothing && set.HasObject {
 		score += 0.03
 	}
 	if set.HasPerson && (set.HasBody || set.HasObject || set.HasClothing) {
@@ -1349,7 +1348,6 @@ func computeHighlightRatingWithUsername(segments []aiSegmentMeta, durationSec fl
 	var totalWeighted float64
 
 	var positionEffectiveWeighted float64
-	var contextEffectiveWeighted float64
 	var peakQuality float64
 
 	var longest float64
@@ -1391,11 +1389,12 @@ func computeHighlightRatingWithUsername(segments []aiSegmentMeta, durationSec fl
 		totalWeighted += weightedDur
 
 		set := ratingSignalSetFromLabel(s.Label)
-		if set.HasPosition {
-			positionEffectiveWeighted += effectiveWeightedDur
-		} else {
-			contextEffectiveWeighted += effectiveWeightedDur
+		if !set.HasPosition {
+			// Nur Positions-Segmente fließen ins Rating ein.
+			// Kleidung und Objekte wirken über Combo-Segmente (position+clothing etc.).
+			continue
 		}
+		positionEffectiveWeighted += effectiveWeightedDur
 
 		if quality > peakQuality {
 			peakQuality = quality
@@ -1420,29 +1419,22 @@ func computeHighlightRatingWithUsername(segments []aiSegmentMeta, durationSec fl
 	avgConfidence := confSum / float64(n)
 
 	positionDensity := positionEffectiveWeighted / videoMinutes
-	contextDensity := contextEffectiveWeighted / videoMinutes
 
 	peakNorm := ratingSmoothStep(peakQuality)
 	positionDensityNorm := ratingSoftCap(positionDensity, 5.5)
-	contextDensityNorm := ratingSoftCap(contextDensity, 8.5)
 	coverageNorm := ratingSoftCap(weightedCoverageRatio, 0.20)
-	frequencyNorm := ratingSoftCap(segmentsPerMinute, 1.25)
 	longestNorm := ratingSoftCap(longest, 24.0)
 	confNorm := ratingSmoothStep((avgConfidence - 0.30) / 0.65)
 
+	// Positionen dominieren. Kleidung/Objekte fließen durch Combo-Segment-Qualität (peakNorm) ein.
+	// Keine Penalties, keine Caps – die Formel bewertet natürlich nach Positions-Intensität.
+	// Summe = 1.00.
 	raw :=
-		0.32*peakNorm +
-			0.25*positionDensityNorm +
-			0.17*contextDensityNorm +
-			0.12*coverageNorm +
-			0.06*longestNorm +
-			0.04*frequencyNorm +
-			0.04*confNorm
-
-	// Ohne Position darf es trotzdem gut werden, aber nicht automatisch 5 Sterne.
-	if positionEffectiveWeighted <= 0 {
-		raw = math.Min(raw, 0.76)
-	}
+		0.42*positionDensityNorm +
+			0.26*peakNorm +
+			0.18*coverageNorm +
+			0.08*longestNorm +
+			0.06*confNorm
 
 	// Sehr wenig Material nicht überbewerten.
 	if totalFlagged < 5.0 && n <= 1 {
@@ -1477,15 +1469,14 @@ func computeHighlightRatingWithUsername(segments []aiSegmentMeta, durationSec fl
 	r.AvgConfidence = ratingRound(avgConfidence, 3)
 
 	appLogf(
-		"✅ %s rating result score=%.1f stars=%d bonus=%.1f segments=%d flagged=%.2f weighted=%.2f coverage=%.4f weightedCoverage=%.4f longest=%.2f avgConf=%.3f",
+		"✅ %s rating score=%.1f stars=%d bonus=%.1f segs=%d flagged=%.1f posDensity=%.2f coverage=%.4f longest=%.1f avgConf=%.3f",
 		ratingLogSubject(username),
 		r.Score,
 		r.Stars,
 		ratingRound(excellenceBonus*100, 1),
 		r.Segments,
 		r.FlaggedSeconds,
-		r.WeightedFlaggedSeconds,
-		r.CoverageRatio,
+		positionDensity,
 		r.WeightedCoverageRatio,
 		r.LongestSegmentSeconds,
 		r.AvgConfidence,
diff --git a/backend/server.go b/backend/server.go
index 79100da..c5cce95 100644
--- a/backend/server.go
+++ b/backend/server.go
@@ -870,7 +870,7 @@ func main() {
 		appLogln("🔐 TLS Cert:", tlsCertFile())
 		appLogln("🔐 TLS Key: ", tlsKeyFile())
 	} else {
-		appLogln("🌐 HTTP-API aktiv: http://localhost:9999")
+		appLogln("🌐 HTTP-API aktiv: https://l14pbbk95100006.tegdssd.de:9999")
 	}
 
 	handler := withCORS(mux)
diff --git a/backend/tray.go b/backend/tray.go
index 67c4b79..207e556 100644
--- a/backend/tray.go
+++ b/backend/tray.go
@@ -108,7 +108,7 @@ func runTray(onQuit func(), statsFn func() TrayStats) {
 					updateStatus()
 
 				case <-mOpenFrontend.ClickedCh:
-					_ = openBrowser("https://localhost:9999")
+					_ = openBrowser("https://l14pbbk95100006.tegdssd.de:9999")
 
 				case <-mOpenRecordDir.ClickedCh:
 					_ = openFolder(resolveConfiguredDir(getSettings().RecordDir))
diff --git a/frontend/src/components/ui/VideoSplitModal.tsx b/frontend/src/components/ui/VideoSplitModal.tsx
index c2870d7..4733e31 100644
--- a/frontend/src/components/ui/VideoSplitModal.tsx
+++ b/frontend/src/components/ui/VideoSplitModal.tsx
@@ -955,8 +955,19 @@ function segmentVisualKindFromText(value: unknown): SegmentVisualKind {
 
 function segmentVisualKind(seg: Segment): SegmentVisualKind {
   const rawLabel = String(seg.label || '').trim()
-  const prettyLabel = getSegmentLabelText(rawLabel)
 
+  // Combo-Labels werden in Backend-Prioritätsreihenfolge gebaut (position > body > object > clothing).
+  // Jeden Teil einzeln prüfen und das erste Ergebnis zurückgeben — so verhält sich auch RatingOverlay.
+  if (rawLabel.toLowerCase().startsWith('combo:')) {
+    const parts = rawLabel.slice(6).split('+')
+    for (const part of parts) {
+      const kind = segmentVisualKindFromText(part.trim())
+      if (kind !== 'default') return kind
+    }
+    return 'default'
+  }
+
+  const prettyLabel = getSegmentLabelText(rawLabel)
   const values = [rawLabel, prettyLabel]
 
   for (const value of values) {