From 733d2bcc7b7aa51635cb0815e76747f469e7a66b Mon Sep 17 00:00:00 2001
From: Benjamin Admin <benjaminadmin@MacBook-Pro.local>
Date: Wed, 13 May 2026 10:00:45 +0200
Subject: [PATCH] feat(iace): per-category hazard caps for precision
 improvement

Add categoryHazardCap() with ISO 12100-proportional limits:
- mechanical: 3x components (min 15, max 60)
- electrical: 1x components (min 8, max 20)
- secondary (thermal, noise, material): 4-8
- software/IT/organizational: 2-5 (minimal for machinery assessment)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 .../api/handlers/iace_handler_init.go         |  8 +++
 .../api/handlers/iace_handler_init_helpers.go | 61 +++++++++++++++++++
 2 files changed, 69 insertions(+)

diff --git a/ai-compliance-sdk/internal/api/handlers/iace_handler_init.go b/ai-compliance-sdk/internal/api/handlers/iace_handler_init.go
index e255034..daf0368 100644
--- a/ai-compliance-sdk/internal/api/handlers/iace_handler_init.go
+++ b/ai-compliance-sdk/internal/api/handlers/iace_handler_init.go
@@ -159,6 +159,7 @@ func (h *IACEHandler) InitializeProject(c *gin.Context) {
 
 		created := 0
 		seenCatZone := make(map[string]bool)
+		catCount := make(map[string]int)
 		for _, mp := range matchOutput.MatchedPatterns {
 			// Narrative relevance filter: skip patterns whose zone/scenario
 			// mentions machine-specific terms that don't appear in our components
@@ -167,6 +168,12 @@ func (h *IACEHandler) InitializeProject(c *gin.Context) {
 			}
 
 			for _, cat := range mp.HazardCats {
+				// Per-category cap: limit hazards per category based on relevance
+				maxForCat := categoryHazardCap(cat, len(comps))
+				if catCount[cat] >= maxForCat {
+					continue
+				}
+
 				// Dedup by category + normalized zone
 				zoneKey := normalizeZoneKey(mp.ZoneDE)
 				if zoneKey == "" {
@@ -212,6 +219,7 @@ func (h *IACEHandler) InitializeProject(c *gin.Context) {
 				})
 				if cerr == nil {
 					created++
+					catCount[cat]++
 					hazardIDsByCategory[cat] = hz.ID
 				}
 			}
diff --git a/ai-compliance-sdk/internal/api/handlers/iace_handler_init_helpers.go b/ai-compliance-sdk/internal/api/handlers/iace_handler_init_helpers.go
index 0f35569..f32fa7b 100644
--- a/ai-compliance-sdk/internal/api/handlers/iace_handler_init_helpers.go
+++ b/ai-compliance-sdk/internal/api/handlers/iace_handler_init_helpers.go
@@ -260,6 +260,67 @@ func isPatternRelevant(mp iace.PatternMatch, narrative string, compNames []strin
 	return false
 }
 
+// categoryHazardCap returns the maximum number of hazards to generate per category.
+// Caps are based on typical ISO 12100 risk assessment proportions:
+// - Core physical categories (mechanical, electrical): scale with component count
+// - Secondary categories (thermal, noise, material): smaller fixed caps
+// - Software/IT/organizational categories: minimal (these are usually covered by
+//   other standards like IEC 62443, not ISO 12100 machinery risk assessment)
+func categoryHazardCap(cat string, componentCount int) int {
+	// Core machinery hazard categories — scale with complexity
+	switch cat {
+	case "mechanical_hazard":
+		// Typically 1-3 hazards per component (quetschen, scheren, stoss...)
+		cap := componentCount * 3
+		if cap < 15 {
+			cap = 15
+		}
+		if cap > 60 {
+			cap = 60
+		}
+		return cap
+	case "electrical_hazard":
+		// Typically 8-15 for a standard machine
+		cap := componentCount
+		if cap < 8 {
+			cap = 8
+		}
+		if cap > 20 {
+			cap = 20
+		}
+		return cap
+	case "pneumatic_hydraulic":
+		return 8
+	case "thermal_hazard":
+		return 6
+	case "noise_vibration":
+		return 4
+	case "material_environmental":
+		return 6
+	case "ergonomic", "ergonomic_hazard":
+		return 4
+	case "fire_explosion":
+		return 4
+	case "radiation_hazard", "emc_hazard":
+		return 3
+	// Software/IT/organizational — minimal for machinery assessment
+	case "safety_function_failure":
+		return 5
+	case "software_fault":
+		return 3
+	case "configuration_error":
+		return 3
+	case "hmi_error":
+		return 3
+	case "maintenance_hazard":
+		return 4
+	case "mode_confusion":
+		return 2
+	default:
+		return 3
+	}
+}
+
 // normalizeZoneKey reduces a zone string to its core components for better dedup.
 // E.g. "Schaltschrank, Sammelschiene" and "Schaltschrank-Innenraum, Sammelschienen"
 // should dedup to the same key.