merge(admin): FE advisor widget — topic threads, per-question delete/copy, fullscreen

Authority Router re-embedded the query per collection (6x); on dev the embed
endpoint (OVH) is remote so that was 6 round-trips = 7-12s per /retrieve. Embed
once, reuse via ctx across the concurrent per-collection searches.
DetectIntent + ConceptNorms now fold ae/oe/ue/ss so ASCII (Pruefe) and umlaut
(Pruefe) inputs both match.

admin-compliance/app/api/sdk/compliance-advisor/chat/route.ts

@@ -1,34 +1,29 @@
 /**
- * Compliance Advisor Chat API
+ * Compliance Advisor Chat API — Clarity-Gate orchestration.
  *
- * Verbindet das ComplianceAdvisorWidget mit:
- *   1. Multi-Collection-RAG ueber die ai-compliance-sdk (bge-m3) — siehe advisor-rag
- *   2. Strukturierten Controls zum erkannten Thema — buildControlsContext
- *   3. LLM-Kaskade OVH (prod) -> Ollama (Dev) — siehe advisor-llm
- *
- * Laenderspezifische Filterung (DE, AT, CH, EU). Streamt die Antwort als Text.
+ * Consumes the SDK/RAG /retrieve (evidence/visual_evidence/footnotes/clarity) and returns the
+ * FE-facing contract (advisor-clarity-gate-contract):
+ *   - clarify mode  -> short L1 general answer (no RAG) + domain context chips
+ *   - answer  mode  -> L2 answer over the scoped evidence with [n] citation markers
+ * Citations are generated here ([n] -> nth evidence unit). The FE renders ONLY this structured data.
  */
 
 import { NextRequest, NextResponse } from 'next/server'
 import { readSoulFile } from '@/lib/sdk/agents/soul-reader'
-import { buildControlsContext } from '@/lib/sdk/agents/controls-augmentation'
-import { queryAdvisorRAG } from '@/lib/sdk/agents/advisor-rag'
-import { streamAdvisorAnswer, type ChatMessage } from '@/lib/sdk/agents/advisor-llm'
+import { retrieveFull } from '@/lib/sdk/agents/advisor-rag'
+import { completeAdvisorAnswer, streamAdvisorAnswer, type ChatMessage } from '@/lib/sdk/agents/advisor-llm'
+import {
+  buildCitations,
+  isLegacyRequest,
+  mapClarity,
+  mapFootnotes,
+  numberedEvidenceForPrompt,
+  resolveMode,
+} from '@/lib/sdk/advisor/retrieve-mapping'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
 
 type Country = 'DE' | 'AT' | 'CH' | 'EU'
 
-const FALLBACK_SYSTEM_PROMPT = `# Compliance Advisor Agent
-
-## Identitaet
-Du bist der BreakPilot Compliance-Berater. Du hilfst Nutzern des AI Compliance SDK,
-Datenschutz- und Compliance-Fragen in verstaendlicher Sprache zu beantworten.
-
-## Kernprinzipien
-- Quellenbasiert: Verweise auf DSGVO-Artikel, BDSG-Paragraphen
-- Verstaendlich: Einfache, praxisnahe Sprache
-- Ehrlich: Bei Unsicherheit empfehle Rechtsberatung
-- Deutsch als Hauptsprache`
-
 const COUNTRY_LABELS: Record<Country, string> = {
   DE: 'Deutschland',
   AT: 'Oesterreich',
@@ -38,84 +33,153 @@ const COUNTRY_LABELS: Record<Country, string> = {
 
 function countryBlock(c: Country): string {
   const label = COUNTRY_LABELS[c]
-  const nationalLaws =
-    c === 'DE'
-      ? 'BDSG, TDDDG, TKG, UWG'
-      : c === 'AT'
-        ? 'AT DSG, ECG, TKG, KSchG, MedienG'
-        : 'CH DSG, DSV, OR, UWG, FMG'
-  const guidance =
-    c === 'EU'
-      ? 'EU-weiten Fragen: Beziehe dich auf EU-Verordnungen und -Richtlinien'
-      : `${label}: Beziehe nationale Gesetze (${nationalLaws}) mit ein`
   return `\n\n## Laenderspezifische Auskunft
-Der Nutzer hat "${label} (${c})" gewaehlt.
-- Beziehe dich AUSSCHLIESSLICH auf ${c}-Recht + anwendbares EU-Recht
-- Nenne IMMER explizit das Land in deiner Antwort
-- Verwende NIEMALS Gesetze eines anderen Landes
-- Bei ${guidance}`
+Der Nutzer hat "${label} (${c})" gewaehlt. Beziehe dich auf ${c}-Recht + anwendbares EU-Recht und nenne das Land.`
+}
+
+// L1: general knowledge, deliberately NOT grounded (the clarify step precedes the legal retrieval).
+const L1_SYSTEM = `Du bist der BreakPilot Compliance-Berater. Gib eine KURZE, allgemeine Definition/Erklaerung
+des gefragten Begriffs aus Allgemeinwissen — 2 bis 4 Saetze, Markdown, neutral. NENNE KEINE Rechtsquellen,
+Paragraphen, Artikel oder Fundstellen; der Nutzer waehlt anschliessend einen konkreten Kontext, erst dann
+folgen belegte Quellen. Wenn der Begriff in mehreren Bereichen vorkommt, erwaehne das in einem Halbsatz.`
+
+const FALLBACK_SYSTEM = `Du bist der BreakPilot Compliance-Berater. Antworte quellenbasiert, verstaendlich und ehrlich auf Deutsch.`
+
+// Optional audience/tonality guidance (e.g. the workspace's role hint). Kept out of the retrieval
+// `question` on purpose — it only shapes the answer's tone, so it belongs in the system prompt.
+function audienceBlock(audience: string): string {
+  return audience ? `\n\n## Ansprache / Zielgruppe\n${audience}` : ''
+}
+
+function answerSystem(
+  soul: string | null,
+  country: Country | undefined,
+  evidenceBlock: string,
+  withCitations = true,
+  audience = '',
+): string {
+  let s = soul || FALLBACK_SYSTEM
+  if (country) s += countryBlock(country)
+  s += audienceBlock(audience)
+  s += `\n\n## Belegte Evidence (nummeriert — DEINE EINZIGEN Quellen)\n${evidenceBlock || '(keine Evidence gefunden)'}`
+  s += `\n\n## Antwortformat (WICHTIG)
+- Beginne mit einer **Kurzzusammenfassung** (1–2 Saetze, "Kurz gesagt: …"), die den Kern direkt beantwortet.
+- Danach gut gegliedertes Markdown: kurze ## Ueberschriften je THEMA/Aspekt (nicht je Rechtsquelle), Aufzaehlungen, **Fettung** fuer Kernbegriffe.`
+  if (withCitations) {
+    s += `\n- Belege Kernaussagen mit [n], wobei n die NUMMER der Evidence-Quelle oben ist (z. B. [1], [2]).
+- Nenne KEINE Quellen-/Fundstellen-Liste im Fliesstext — die Quellen werden dem Nutzer separat angezeigt.`
+  } else {
+    s += `\n- Nenne Fundstellen nur, wo sie der Antwort dienen (natuerlich im Text, KEIN [n]-Markup).`
+  }
+  s += `\n- Triff KEINE Aussage, die nicht durch die nummerierte Evidence belegt ist; fehlt der Beleg, sage das offen.`
+  return s
+}
+
+// Prior thread turns for contextual follow-ups. Validated + bounded (last 8 turns ~ 4 Q&A).
+function parseHistory(raw: unknown): ChatMessage[] {
+  if (!Array.isArray(raw)) return []
+  const turns: ChatMessage[] = []
+  for (const t of raw) {
+    if (!t || typeof t !== 'object') continue
+    const role = (t as { role?: unknown }).role
+    const content = (t as { content?: unknown }).content
+    if ((role === 'user' || role === 'assistant') && typeof content === 'string' && content.trim()) {
+      turns.push({ role, content })
+    }
+  }
+  return turns.slice(-8)
 }
 
 export async function POST(request: NextRequest) {
   try {
     const body = await request.json()
-    const { message, history = [], currentStep = 'default', country } = body
-
-    if (!message || typeof message !== 'string') {
-      return NextResponse.json({ error: 'Message is required' }, { status: 400 })
-    }
-
-    const validCountry = (['DE', 'AT', 'CH', 'EU'] as const).includes(country)
-      ? (country as Country)
+    const question = String(body.question ?? body.message ?? '').trim()
+    const context: string | null = body.context ?? null
+    const audience = typeof body.audience === 'string' ? body.audience.trim() : ''
+    const history = parseHistory(body.history)
+    const country = (['DE', 'AT', 'CH', 'EU'] as const).includes(body.country)
+      ? (body.country as Country)
       : undefined
 
-    // 1. RAG (ai-sdk, bge-m3) + strukturierte Controls zum Thema — beide parallel
-    const [ragContext, controlsContext] = await Promise.all([
-      queryAdvisorRAG(message),
-      buildControlsContext(message),
+    if (!question) {
+      return NextResponse.json({ error: 'Question is required' }, { status: 400 })
+    }
+
+    const retrieved = await retrieveFull(question, context)
+
+    // Backward-compat: legacy consumers (breakpilot-workspace) send {message} and read a plain-text
+    // stream. Serve the L2 answer streamed (clean prose, no [n]); no clarify gate, no JSON.
+    if (isLegacyRequest(body)) {
+      const legacyEvidence = retrieved.evidence ?? []
+      const legacySoul = await readSoulFile('compliance-advisor')
+      const legacyStream = await streamAdvisorAnswer([
+        { role: 'system', content: answerSystem(legacySoul, country, numberedEvidenceForPrompt(legacyEvidence), false, audience) },
+        ...history,
+        { role: 'user', content: question },
       ])
-
-    // 2. System-Prompt zusammenbauen
-    const soulPrompt = await readSoulFile('compliance-advisor')
-    let systemContent = soulPrompt || FALLBACK_SYSTEM_PROMPT
-    if (validCountry) systemContent += countryBlock(validCountry)
-    if (ragContext) {
-      systemContent += `\n\n## Relevanter Kontext aus dem RAG-System (deine EINZIGEN Rechtsquellen)\n\nDies sind deine einzigen zulaessigen Rechtsquellen. Triff keine konkrete Rechtsaussage (Zahl, Frist, Schwelle, Pflicht, Fundstelle), die nicht hier oder im Controls-Block belegt ist — sonst sage offen, dass du sie aus deinen Quellen nicht belegen kannst. Verweise in deiner Antwort auf die jeweilige Quelle:\n\n${ragContext}`
+      if (!legacyStream) {
+        return NextResponse.json({ error: 'LLM nicht erreichbar.' }, { status: 502 })
       }
-    if (controlsContext) systemContent += `\n\n${controlsContext}`
-    systemContent += `\n\n## Aktueller SDK-Schritt\nDer Nutzer befindet sich im SDK-Schritt: ${currentStep}`
-
-    // 3. Nachrichten (History auf die letzten 6 begrenzen)
-    const messages: ChatMessage[] = [
-      { role: 'system', content: systemContent },
-      ...history.slice(-6).map((h: { role: string; content: string }) => ({
-        role: h.role === 'user' ? 'user' : 'assistant',
-        content: h.content,
-      })),
-      { role: 'user', content: message },
-    ]
-
-    // 4. LLM-Kaskade -> Plain-Text-Stream
-    const stream = await streamAdvisorAnswer(messages)
-    if (!stream) {
-      return NextResponse.json(
-        { error: 'LLM nicht erreichbar. Weder OVH/LiteLLM noch Ollama haben geantwortet.' },
-        { status: 502 },
-      )
-    }
-
-    return new NextResponse(stream, {
+      return new NextResponse(legacyStream, {
         headers: {
           'Content-Type': 'text/plain; charset=utf-8',
           'Cache-Control': 'no-cache',
-        Connection: 'keep-alive',
+          'X-Advisor-Format': 'legacy-stream',
         },
       })
+    }
+
+    const mode = resolveMode(retrieved.clarity?.mode, !!context)
+
+    if (mode === 'clarify') {
+      const general = await completeAdvisorAnswer([
+        { role: 'system', content: L1_SYSTEM + audienceBlock(audience) },
+        { role: 'user', content: question },
+      ])
+      if (general === null) {
+        return NextResponse.json({ error: 'LLM nicht erreichbar.' }, { status: 502 })
+      }
+      const resp: AdvisorResponse = {
+        mode: 'clarify',
+        question,
+        clarity: mapClarity(retrieved.clarity, 'clarify'),
+        general_answer: general,
+        answer: null,
+        scoped_query: null,
+        evidence: [],
+        citations: [],
+        visual_evidence: [],
+        footnotes: [],
+      }
+      return NextResponse.json(resp)
+    }
+
+    const evidence = retrieved.evidence ?? []
+    const soul = await readSoulFile('compliance-advisor')
+    const messages: ChatMessage[] = [
+      { role: 'system', content: answerSystem(soul, country, numberedEvidenceForPrompt(evidence), true, audience) },
+      ...history,
+      { role: 'user', content: question },
+    ]
+    const answer = await completeAdvisorAnswer(messages)
+    if (answer === null) {
+      return NextResponse.json({ error: 'LLM nicht erreichbar.' }, { status: 502 })
+    }
+    const resp: AdvisorResponse = {
+      mode: 'answer',
+      question,
+      clarity: mapClarity(retrieved.clarity, 'answer'),
+      general_answer: null,
+      answer,
+      scoped_query: context,
+      evidence,
+      citations: buildCitations(evidence),
+      visual_evidence: retrieved.visual_evidence ?? [],
+      footnotes: mapFootnotes(retrieved.footnotes),
+    }
+    return NextResponse.json(resp)
   } catch (error) {
     console.error('Compliance advisor chat error:', error)
-    return NextResponse.json(
-      { error: 'Verbindung zum LLM fehlgeschlagen.' },
-      { status: 503 },
-    )
+    return NextResponse.json({ error: 'Verbindung zum Advisor fehlgeschlagen.' }, { status: 503 })
   }
 }

admin-compliance/components/sdk/ComplianceAdvisorParts.tsx

@@ -1,131 +0,0 @@
-'use client'
-
-// =============================================================================
-// ComplianceAdvisorWidget — shared constants and sub-components
-// =============================================================================
-
-// =============================================================================
-// EXAMPLE QUESTIONS
-// =============================================================================
-
-export const EXAMPLE_QUESTIONS: Record<string, string[]> = {
-  vvt: [
-    'Was ist ein Verarbeitungsverzeichnis?',
-    'Welche Informationen muss ich erfassen?',
-    'Wie dokumentiere ich die Rechtsgrundlage?',
-  ],
-  'compliance-scope': [
-    'Was bedeutet L3?',
-    'Wann brauche ich eine DSFA?',
-    'Was ist der Unterschied zwischen L2 und L3?',
-  ],
-  tom: [
-    'Was sind TOM?',
-    'Welche Massnahmen sind erforderlich?',
-    'Wie dokumentiere ich Verschluesselung?',
-  ],
-  dsfa: [
-    'Was ist eine DSFA?',
-    'Wann ist eine DSFA verpflichtend?',
-    'Wie bewerte ich Risiken?',
-  ],
-  loeschfristen: [
-    'Wie definiere ich Loeschfristen?',
-    'Was ist der Unterschied zwischen Loeschpflicht und Aufbewahrungspflicht?',
-    'Wann muss ich Daten loeschen?',
-  ],
-  default: [
-    'Wie starte ich mit dem SDK?',
-    'Was ist der erste Schritt?',
-    'Welche Compliance-Anforderungen gelten fuer KI-Systeme?',
-  ],
-}
-
-// =============================================================================
-// TYPES
-// =============================================================================
-
-export interface Message {
-  id: string
-  role: 'user' | 'agent'
-  content: string
-  timestamp: Date
-}
-
-// =============================================================================
-// EmptyState — shown when no messages yet
-// =============================================================================
-
-interface EmptyStateProps {
-  exampleQuestions: string[]
-  onExampleClick: (question: string) => void
-}
-
-export function AdvisorEmptyState({ exampleQuestions, onExampleClick }: EmptyStateProps) {
-  return (
-    <div className="text-center py-8">
-      <div className="w-16 h-16 bg-purple-100 rounded-full flex items-center justify-center mx-auto mb-4">
-        <svg className="w-8 h-8 text-purple-600" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-          <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M8 10h.01M12 10h.01M16 10h.01M9 16H5a2 2 0 01-2-2V6a2 2 0 012-2h14a2 2 0 012 2v8a2 2 0 01-2 2h-5l-5 5v-5z" />
-        </svg>
-      </div>
-      <h3 className="text-sm font-medium text-gray-900 mb-2">Willkommen beim Compliance Advisor</h3>
-      <p className="text-xs text-gray-500 mb-4">Stellen Sie Fragen zu DSGVO, KI-Verordnung und mehr.</p>
-      <div className="text-left space-y-2">
-        <p className="text-xs font-medium text-gray-700 mb-2">Beispielfragen:</p>
-        {exampleQuestions.map((question, idx) => (
-          <button
-            key={idx}
-            onClick={() => onExampleClick(question)}
-            className="w-full text-left px-3 py-2 text-xs bg-white hover:bg-purple-50 border border-gray-200 rounded-lg transition-colors text-gray-700"
-          >
-            {question}
-          </button>
-        ))}
-      </div>
-    </div>
-  )
-}
-
-// =============================================================================
-// MessageList — renders messages + typing indicator
-// =============================================================================
-
-interface MessageListProps {
-  messages: Message[]
-  isTyping: boolean
-  messagesEndRef: React.RefObject<HTMLDivElement | null>
-}
-
-export function AdvisorMessageList({ messages, isTyping, messagesEndRef }: MessageListProps) {
-  return (
-    <>
-      {messages.map((message) => (
-        <div key={message.id} className={`flex ${message.role === 'user' ? 'justify-end' : 'justify-start'}`}>
-          <div className={`max-w-[80%] rounded-lg px-3 py-2 ${message.role === 'user' ? 'bg-indigo-600 text-white' : 'bg-white border border-gray-200 text-gray-800'}`}>
-            <p className={`text-sm ${message.role === 'agent' ? 'whitespace-pre-wrap' : ''}`}>
-              {message.content || (message.role === 'agent' && isTyping ? '' : message.content)}
-            </p>
-            <p className={`text-xs mt-1 ${message.role === 'user' ? 'text-indigo-200' : 'text-gray-400'}`}>
-              {message.timestamp.toLocaleTimeString('de-DE', { hour: '2-digit', minute: '2-digit' })}
-            </p>
-          </div>
-        </div>
-      ))}
-
-      {isTyping && (
-        <div className="flex justify-start">
-          <div className="bg-white border border-gray-200 rounded-lg px-3 py-2">
-            <div className="flex space-x-1">
-              <div className="w-2 h-2 bg-gray-400 rounded-full animate-bounce" />
-              <div className="w-2 h-2 bg-gray-400 rounded-full animate-bounce" style={{ animationDelay: '0.1s' }} />
-              <div className="w-2 h-2 bg-gray-400 rounded-full animate-bounce" style={{ animationDelay: '0.2s' }} />
-            </div>
-          </div>
-        </div>
-      )}
-
-      <div ref={messagesEndRef as React.RefObject<HTMLDivElement>} />
-    </>
-  )
-}

admin-compliance/components/sdk/ComplianceAdvisorWidget.tsx

@@ -1,198 +1,78 @@
 'use client'
 
-import { useState, useEffect, useRef, useCallback } from 'react'
-import { EXAMPLE_QUESTIONS, AdvisorEmptyState, AdvisorMessageList, type Message } from './ComplianceAdvisorParts'
-
-// =============================================================================
-// TYPES
-// =============================================================================
+import { useCallback, useState } from 'react'
+import {
+  Check,
+  Expand,
+  Loader2,
+  Mail,
+  Maximize2,
+  MessagesSquare,
+  Minimize2,
+  Plus,
+  Send,
+  Shrink,
+  Square,
+  X,
+} from 'lucide-react'
+import { EXAMPLE_QUESTIONS } from './advisor/EmptyState'
+import { EvidenceWorkspace } from './advisor/EvidenceWorkspace'
+import { useAdvisorCase } from './advisor/useAdvisorCase'
+import { useAdvisorEmail } from './advisor/useAdvisorEmail'
 
 interface ComplianceAdvisorWidgetProps {
   currentStep?: string
 }
 
 type Country = 'DE' | 'AT' | 'CH' | 'EU'
+const COUNTRIES: Country[] = ['DE', 'AT', 'CH', 'EU']
 
-const COUNTRIES: { code: Country; label: string }[] = [
-  { code: 'DE', label: 'DE' },
-  { code: 'AT', label: 'AT' },
-  { code: 'CH', label: 'CH' },
-  { code: 'EU', label: 'EU' },
-]
-
-// =============================================================================
-// COMPONENT
-// =============================================================================
+type View = 'compact' | 'panel' | 'fullscreen'
+const SIZE: Record<View, string> = {
+  compact: 'bottom-6 right-6 h-[560px] w-[420px] rounded-2xl',
+  panel: 'bottom-6 right-6 h-[85vh] w-[960px] rounded-2xl',
+  fullscreen: 'inset-0 h-full w-full',
+}
 
+/**
+ * Compliance Advisor — a floating Case Workspace on every SDK page (compact / panel / fullscreen).
+ * Renders ONLY structured SDK data (clarify/answer contract); it never parses the answer text.
+ * See memory: advisor-evidence-workspace-no-parse, advisor-clarity-gate-contract.
+ */
 export function ComplianceAdvisorWidget({ currentStep = 'default' }: ComplianceAdvisorWidgetProps) {
   const [isOpen, setIsOpen] = useState(false)
-  const [isExpanded, setIsExpanded] = useState(false)
-  const [messages, setMessages] = useState<Message[]>([])
+  const [view, setView] = useState<View>('compact')
   const [inputValue, setInputValue] = useState('')
-  const [isTyping, setIsTyping] = useState(false)
-  const [selectedCountry, setSelectedCountry] = useState<Country>('DE')
-  const messagesEndRef = useRef<HTMLDivElement>(null)
-  const abortControllerRef = useRef<AbortController | null>(null)
+  const [country, setCountry] = useState<Country>('DE')
 
+  const { cases, threads, busy, activeCaseId, ask, newTopic, selectContext, selectCase, remove, stop } =
+    useAdvisorCase({ currentStep, country })
+  const email = useAdvisorEmail(cases, country, currentStep)
   const exampleQuestions = EXAMPLE_QUESTIONS[currentStep] || EXAMPLE_QUESTIONS.default
+  const expanded = view !== 'compact'
 
-  useEffect(() => {
-    messagesEndRef.current?.scrollIntoView({ behavior: 'smooth' })
-  }, [messages])
-
-  useEffect(() => {
-    return () => {
-      abortControllerRef.current?.abort()
-    }
-  }, [])
-
-  const handleSendMessage = useCallback(
-    async (content: string) => {
-      if (!content.trim() || isTyping) return
-
-      const userMessage: Message = {
-        id: `msg-${Date.now()}`,
-        role: 'user',
-        content: content.trim(),
-        timestamp: new Date(),
-      }
-
-      setMessages((prev) => [...prev, userMessage])
+  const submit = useCallback(
+    (q: string) => {
+      if (!q.trim() || busy) return
       setInputValue('')
-      setIsTyping(true)
-
-      const agentMessageId = `msg-${Date.now()}-agent`
-      abortControllerRef.current = new AbortController()
-
-      try {
-        const history = messages.map((m) => ({
-          role: m.role === 'user' ? 'user' : 'assistant',
-          content: m.content,
-        }))
-
-        const response = await fetch('/api/sdk/compliance-advisor/chat', {
-          method: 'POST',
-          headers: { 'Content-Type': 'application/json' },
-          body: JSON.stringify({
-            message: content.trim(),
-            history,
-            currentStep,
-            country: selectedCountry,
-          }),
-          signal: abortControllerRef.current.signal,
-        })
-
-        if (!response.ok) {
-          const errorData = await response.json().catch(() => ({ error: 'Unbekannter Fehler' }))
-          throw new Error(errorData.error || `Server-Fehler (${response.status})`)
-        }
-
-        setMessages((prev) => [
-          ...prev,
-          { id: agentMessageId, role: 'agent', content: '', timestamp: new Date() },
-        ])
-
-        const reader = response.body!.getReader()
-        const decoder = new TextDecoder()
-        let accumulated = ''
-
-        while (true) {
-          const { done, value } = await reader.read()
-          if (done) break
-
-          accumulated += decoder.decode(value, { stream: true })
-          const currentText = accumulated
-          setMessages((prev) =>
-            prev.map((m) => (m.id === agentMessageId ? { ...m, content: currentText } : m))
-          )
-          requestAnimationFrame(() => {
-            messagesEndRef.current?.scrollIntoView({ behavior: 'smooth' })
-          })
-        }
-
-        setIsTyping(false)
-      } catch (error) {
-        if ((error as Error).name === 'AbortError') {
-          setIsTyping(false)
-          return
-        }
-
-        const errorMessage = error instanceof Error ? error.message : 'Verbindung fehlgeschlagen'
-        setMessages((prev) => {
-          const hasAgent = prev.some((m) => m.id === agentMessageId)
-          if (hasAgent) {
-            return prev.map((m) =>
-              m.id === agentMessageId ? { ...m, content: `Fehler: ${errorMessage}` } : m
-            )
-          }
-          return [
-            ...prev,
-            { id: agentMessageId, role: 'agent' as const, content: `Fehler: ${errorMessage}`, timestamp: new Date() },
-          ]
-        })
-        setIsTyping(false)
-      }
+      ask(q)
     },
-    [isTyping, messages, currentStep, selectedCountry]
+    [busy, ask],
   )
 
-  const handleStopGeneration = useCallback(() => {
-    abortControllerRef.current?.abort()
-    setIsTyping(false)
-  }, [])
+  const submitNewTopic = useCallback(
+    (q: string) => {
+      if (!q.trim() || busy) return
+      setInputValue('')
+      newTopic(q)
+    },
+    [busy, newTopic],
+  )
 
-  const [emailSending, setEmailSending] = useState(false)
-  const [emailSent, setEmailSent] = useState(false)
-
-  const handleSendAsEmail = useCallback(async () => {
-    if (messages.length === 0 || emailSending) return
-    setEmailSending(true)
-    try {
-      // Build HTML from chat messages
-      const qaPairs = messages.reduce<{ q: string; a: string }[]>((acc, m, i) => {
-        if (m.role === 'user') {
-          const next = messages[i + 1]
-          acc.push({ q: m.content, a: next?.role === 'agent' ? next.content : '(keine Antwort)' })
-        }
-        return acc
-      }, [])
-
-      const qaHtml = qaPairs.map(({ q, a }) =>
-        `<div style="margin-bottom:16px;"><p style="font-weight:600;color:#1e293b;">Frage: ${q}</p><p style="color:#475569;white-space:pre-wrap;">${a}</p></div>`
-      ).join('')
-
-      const bodyHtml = `
-        <h2 style="color:#1e293b;">Compliance Advisor — Beratungsprotokoll</h2>
-        <p style="color:#64748b;font-size:13px;">Datum: ${new Date().toLocaleString('de-DE')} | Land: ${selectedCountry} | Kontext: ${currentStep}</p>
-        <hr style="border-color:#e2e8f0;margin:16px 0;">
-        ${qaHtml}
-        <hr style="border-color:#e2e8f0;margin:16px 0;">
-        <p style="color:#94a3b8;font-size:11px;">Automatisch erstellt vom BreakPilot Compliance Advisor (Qwen)</p>
-      `
-
-      await fetch('/api/sdk/v1/agent/notify', {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-          recipient: 'dsb@breakpilot.local',
-          subject: `Compliance Advisor — ${qaPairs.length} Fragen (${currentStep})`,
-          body_html: bodyHtml,
-          role: 'Datenschutzbeauftragter',
-        }),
-      })
-      setEmailSent(true)
-      setTimeout(() => setEmailSent(false), 3000)
-    } catch (e) {
-      console.error('Email send failed:', e)
-    } finally {
-      setEmailSending(false)
-    }
-  }, [messages, emailSending, selectedCountry, currentStep])
-
-  const handleKeyDown = (e: React.KeyboardEvent) => {
+  const onKeyDown = (e: React.KeyboardEvent) => {
     if (e.key === 'Enter' && !e.shiftKey) {
       e.preventDefault()
-      handleSendMessage(inputValue)
+      submit(inputValue)
     }
   }
 
@@ -200,137 +80,133 @@ export function ComplianceAdvisorWidget({ currentStep = 'default' }: ComplianceA
     return (
       <button
         onClick={() => setIsOpen(true)}
-        className="fixed bottom-6 right-[5.5rem] w-14 h-14 bg-indigo-600 hover:bg-indigo-700 text-white rounded-full shadow-lg flex items-center justify-center transition-all duration-200 hover:scale-110 z-50"
+        className="fixed bottom-6 right-[5.5rem] z-50 flex h-14 w-14 items-center justify-center rounded-full bg-indigo-600 text-white shadow-lg transition-all duration-200 hover:scale-110 hover:bg-indigo-700"
         aria-label="Compliance Advisor oeffnen"
       >
-        <svg className="w-6 h-6" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-          <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M8 10h.01M12 10h.01M16 10h.01M9 16H5a2 2 0 01-2-2V6a2 2 0 012-2h14a2 2 0 012 2v8a2 2 0 01-2 2h-5l-5 5v-5z" />
-        </svg>
+        <MessagesSquare className="h-6 w-6" />
       </button>
     )
   }
 
+  const headRound = view === 'fullscreen' ? '' : 'rounded-t-2xl'
+  const footRound = view === 'fullscreen' ? '' : 'rounded-b-2xl'
+
   return (
-    <div className={`fixed bottom-6 right-6 ${isExpanded ? 'w-[700px] h-[80vh]' : 'w-[400px] h-[500px]'} max-h-screen bg-white rounded-2xl shadow-2xl flex flex-col z-50 border border-gray-200 transition-all duration-200`}>
-      {/* Header */}
-      <div className="bg-gradient-to-r from-purple-600 to-indigo-600 text-white px-4 py-3 rounded-t-2xl flex items-center justify-between">
+    <div
+      className={`fixed z-50 flex max-h-screen flex-col border border-gray-200 bg-white shadow-2xl transition-all duration-200 ${SIZE[view]}`}
+    >
+      <div className={`flex items-center justify-between bg-gradient-to-r from-purple-600 to-indigo-600 px-4 py-3 text-white ${headRound}`}>
         <div className="flex items-center gap-2">
-          <div className="w-8 h-8 bg-white/20 rounded-full flex items-center justify-center">
-            <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-              <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M9.663 17h4.673M12 3v1m6.364 1.636l-.707.707M21 12h-1M4 12H3m3.343-5.657l-.707-.707m2.828 9.9a5 5 0 117.072 0l-.548.547A3.374 3.374 0 0014 18.469V19a2 2 0 11-4 0v-.531c0-.895-.356-1.754-.988-2.386l-.548-.547z" />
-            </svg>
+          <div className="flex h-8 w-8 items-center justify-center rounded-full bg-white/20">
+            <MessagesSquare className="h-5 w-5" />
           </div>
           <div>
-            <div className="font-semibold text-sm">Compliance Advisor</div>
-            <div className="flex items-center gap-1 mt-0.5">
-              {COUNTRIES.map(({ code, label }) => (
+            <div className="text-sm font-semibold">Compliance Advisor</div>
+            <div className="mt-0.5 flex items-center gap-1">
+              {COUNTRIES.map((c) => (
                 <button
-                  key={code}
-                  onClick={() => setSelectedCountry(code)}
-                  className={`px-1.5 py-0.5 text-[10px] font-medium rounded transition-colors ${selectedCountry === code ? 'bg-white text-indigo-700' : 'bg-white/15 text-white/80 hover:bg-white/25'}`}
+                  key={c}
+                  onClick={() => setCountry(c)}
+                  className={`rounded px-1.5 py-0.5 text-[10px] font-medium transition-colors ${
+                    country === c ? 'bg-white text-indigo-700' : 'bg-white/15 text-white/80 hover:bg-white/25'
+                  }`}
                 >
-                  {label}
+                  {c}
                 </button>
               ))}
             </div>
           </div>
         </div>
         <div className="flex items-center gap-1">
-          {/* Send as Email */}
-          {messages.length > 0 && (
+          {cases.length > 0 && (
             <button
-              onClick={handleSendAsEmail}
-              disabled={emailSending}
-              className={`text-white/80 hover:text-white transition-colors ${emailSent ? 'text-green-300' : ''}`}
+              onClick={email.send}
+              disabled={email.sending}
+              className={`text-white/80 transition-colors hover:text-white ${email.sent ? 'text-green-300' : ''}`}
+              title={email.sent ? 'Email gesendet!' : 'Beratungsprotokoll als Email senden'}
               aria-label="Als Email an DSB senden"
-              title={emailSent ? 'Email gesendet!' : 'Beratungsprotokoll als Email senden'}
             >
-              {emailSent ? (
-                <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-                  <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M5 13l4 4L19 7" />
-                </svg>
-              ) : emailSending ? (
-                <svg className="w-5 h-5 animate-spin" fill="none" viewBox="0 0 24 24">
-                  <circle className="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" strokeWidth="4" />
-                  <path className="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8V0C5.373 0 0 5.373 0 12h4z" />
-                </svg>
-              ) : (
-                <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-                  <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M3 8l7.89 5.26a2 2 0 002.22 0L21 8M5 19h14a2 2 0 002-2V7a2 2 0 00-2-2H5a2 2 0 00-2 2v10a2 2 0 002 2z" />
-                </svg>
+              {email.sent ? <Check className="h-5 w-5" /> : email.sending ? <Loader2 className="h-5 w-5 animate-spin" /> : <Mail className="h-5 w-5" />}
+            </button>
           )}
+          {view !== 'fullscreen' && (
+            <button
+              onClick={() => setView((v) => (v === 'compact' ? 'panel' : 'compact'))}
+              className="text-white/80 transition-colors hover:text-white"
+              aria-label={view === 'compact' ? 'Vergroessern' : 'Verkleinern'}
+            >
+              {view === 'compact' ? <Maximize2 className="h-5 w-5" /> : <Minimize2 className="h-5 w-5" />}
             </button>
           )}
           <button
-            onClick={() => setIsExpanded(!isExpanded)}
-            className="text-white/80 hover:text-white transition-colors"
-            aria-label={isExpanded ? 'Verkleinern' : 'Vergroessern'}
+            onClick={() => setView((v) => (v === 'fullscreen' ? 'panel' : 'fullscreen'))}
+            className="text-white/80 transition-colors hover:text-white"
+            aria-label={view === 'fullscreen' ? 'Vollbild verlassen' : 'Vollbild'}
+            title={view === 'fullscreen' ? 'Vollbild verlassen' : 'Vollbild'}
           >
-            <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-              {isExpanded ? (
-                <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M9 9L4 4m0 0v4m0-4h4m6 6l5 5m0 0v-4m0 4h-4" />
-              ) : (
-                <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M4 8V4m0 0h4M4 4l5 5m11-1V4m0 0h-4m4 0l-5 5M4 16v4m0 0h4m-4 0l5-5m11 5v-4m0 4h-4m4 0l-5-5" />
-              )}
-            </svg>
+            {view === 'fullscreen' ? <Shrink className="h-5 w-5" /> : <Expand className="h-5 w-5" />}
           </button>
-          <button onClick={() => setIsOpen(false)} className="text-white/80 hover:text-white transition-colors" aria-label="Schliessen">
-            <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-              <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M6 18L18 6M6 6l12 12" />
-            </svg>
+          <button
+            onClick={() => setIsOpen(false)}
+            className="text-white/80 transition-colors hover:text-white"
+            aria-label="Schliessen"
+          >
+            <X className="h-5 w-5" />
           </button>
         </div>
       </div>
 
-      {/* Messages Area */}
-      <div className="flex-1 overflow-y-auto p-4 space-y-4 bg-gray-50">
-        {messages.length === 0 ? (
-          <AdvisorEmptyState
+      <EvidenceWorkspace
+        cases={cases}
+        threads={threads}
+        expanded={expanded}
+        busy={busy}
+        activeCaseId={activeCaseId}
         exampleQuestions={exampleQuestions}
-            onExampleClick={(q) => handleSendMessage(q)}
+        onExample={submit}
+        onSelectContext={selectContext}
+        onSelectCase={selectCase}
+        onRemove={remove}
       />
-        ) : (
-          <AdvisorMessageList
-            messages={messages}
-            isTyping={isTyping}
-            messagesEndRef={messagesEndRef}
-          />
-        )}
-      </div>
 
-      {/* Input Area */}
-      <div className="border-t border-gray-200 p-3 bg-white rounded-b-2xl">
+      <div className={`border-t border-gray-200 bg-white p-3 ${footRound}`}>
         <div className="flex gap-2">
           <input
             type="text"
             value={inputValue}
             onChange={(e) => setInputValue(e.target.value)}
-            onKeyDown={handleKeyDown}
-            placeholder="Frage eingeben..."
-            disabled={isTyping}
-            className="flex-1 px-3 py-2 text-sm border border-gray-300 rounded-lg focus:outline-none focus:ring-2 focus:ring-purple-500 focus:border-transparent disabled:opacity-50"
+            onKeyDown={onKeyDown}
+            placeholder={cases.length > 0 ? 'Folgefrage eingeben...' : 'Frage eingeben...'}
+            disabled={busy}
+            className="flex-1 rounded-lg border border-gray-300 px-3 py-2 text-sm focus:border-transparent focus:outline-none focus:ring-2 focus:ring-purple-500 disabled:opacity-50"
           />
-          {isTyping ? (
-            <button
-              onClick={handleStopGeneration}
-              className="px-4 py-2 bg-red-500 text-white rounded-lg hover:bg-red-600 transition-colors"
-              title="Generierung stoppen"
-            >
-              <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-                <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M6 6h12v12H6z" />
-              </svg>
+          {busy ? (
+            <button onClick={stop} className="rounded-lg bg-red-500 px-4 py-2 text-white transition-colors hover:bg-red-600" title="Abbrechen">
+              <Square className="h-5 w-5" />
             </button>
           ) : (
+            <>
+              {cases.length > 0 && (
                 <button
-              onClick={() => handleSendMessage(inputValue)}
+                  onClick={() => submitNewTopic(inputValue)}
                   disabled={!inputValue.trim()}
-              className="px-4 py-2 bg-indigo-600 text-white rounded-lg hover:bg-indigo-700 disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+                  className="rounded-lg border border-gray-300 px-3 py-2 text-gray-600 transition-colors hover:bg-gray-50 disabled:cursor-not-allowed disabled:opacity-50"
+                  title="Als neues Thema stellen"
+                  aria-label="Neues Thema"
                 >
-              <svg className="w-5 h-5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
-                <path strokeLinecap="round" strokeLinejoin="round" strokeWidth={2} d="M12 19l9 2-9-18-9 18 9-2zm0 0v-8" />
-              </svg>
+                  <Plus className="h-5 w-5" />
                 </button>
               )}
+              <button
+                onClick={() => submit(inputValue)}
+                disabled={!inputValue.trim()}
+                className="rounded-lg bg-indigo-600 px-4 py-2 text-white transition-colors hover:bg-indigo-700 disabled:cursor-not-allowed disabled:opacity-50"
+                title={cases.length > 0 ? 'Folgefrage senden' : 'Frage senden'}
+              >
+                <Send className="h-5 w-5" />
+              </button>
+            </>
+          )}
         </div>
       </div>
     </div>

admin-compliance/components/sdk/advisor/CaseView.test.tsx

@@ -0,0 +1,68 @@
+import { describe, it, expect, vi } from 'vitest'
+import { render, fireEvent } from '@testing-library/react'
+import { CaseView } from './CaseView'
+import type { AdvisorCase } from './useAdvisorCase'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
+
+const clarify: AdvisorResponse = {
+  mode: 'clarify',
+  question: 'Was ist PDCA?',
+  clarity: {
+    is_underspecified: true,
+    concentration: 0.38,
+    suggested_contexts: [
+      { id: 'datenschutz', label: 'Datenschutz' },
+      { id: 'qm', label: 'Qualitätsmanagement' },
+    ],
+  },
+  general_answer: 'PDCA steht für **Plan-Do-Check-Act**.',
+  answer: null,
+  evidence: [],
+  citations: [],
+  visual_evidence: [],
+  footnotes: [],
+}
+
+const answer: AdvisorResponse = {
+  mode: 'answer',
+  question: 'PDCA im Datenschutz?',
+  clarity: { is_underspecified: false, dominant_context: 'datenschutz', concentration: 0.88 },
+  answer: 'Der DSM-Zyklus [1] beschreibt den Ablauf.',
+  evidence: [
+    { evidence_id: 'e1', document: 'DSK Sdm B41', section: 'Art. 5', paragraph: 'Abs. 2', snippet: 'x' },
+  ],
+  citations: [
+    { citation_id: 'c1', evidence_id: 'e1', document: 'DSK Sdm B41', section: 'Art. 5', paragraph: 'Abs. 2' },
+  ],
+  visual_evidence: [
+    { visual_id: 'v1', visual_type: 'flowchart', caption: 'PDCA-Zyklus', document: 'DSK SDM', vision_summary: 's' },
+  ],
+  footnotes: [],
+}
+
+function mk(response: AdvisorResponse): AdvisorCase {
+  return { id: 'case1', threadId: 'thread1', question: response.question, response, selectedContext: null, status: 'done' }
+}
+
+describe('CaseView — clarify mode', () => {
+  it('renders the L1 general answer + context chips and fires onSelectContext', () => {
+    const onSel = vi.fn()
+    const { container, getByText } = render(
+      <CaseView c={mk(clarify)} busy={false} onSelectContext={onSel} />,
+    )
+    expect(container.textContent).toContain('Plan-Do-Check-Act')
+    expect(container.textContent).toContain('Allgemeine Definition')
+    fireEvent.click(getByText('Datenschutz'))
+    expect(onSel).toHaveBeenCalledWith('datenschutz')
+  })
+})
+
+describe('CaseView — answer mode', () => {
+  it('renders answer with a clickable [n] citation, grouped evidence (friendly name), and visual', () => {
+    const { container } = render(<CaseView c={mk(answer)} busy={false} onSelectContext={() => {}} />)
+    expect(container.textContent).toContain('DSM-Zyklus')
+    expect(container.querySelector('button[title="Beleg 1 anzeigen"]')).not.toBeNull()
+    expect(container.textContent).toContain('DSK Standard-Datenschutzmodell')
+    expect(container.textContent).toContain('PDCA-Zyklus')
+  })
+})

admin-compliance/components/sdk/advisor/CaseView.tsx

@@ -0,0 +1,107 @@
+'use client'
+
+import { Check, Copy, Trash2 } from 'lucide-react'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
+import { formatCaseForCopy } from '@/lib/sdk/advisor/copy'
+import type { AdvisorCase } from './useAdvisorCase'
+import { ClarifyView } from './ClarifyView'
+import { EvidenceSummary } from './EvidenceSummary'
+import { EvidencePane } from './EvidencePane'
+import { VisualEvidencePane } from './VisualEvidencePane'
+import { FootnotesPane } from './FootnotesPane'
+import { Markdown } from './Markdown'
+import { useCitationHighlight } from './useCitationHighlight'
+import { useClipboard } from './useClipboard'
+
+export function LoadingDots() {
+  return (
+    <div className="flex space-x-1 px-1 py-2" aria-label="Antwort wird erstellt">
+      <span className="h-2 w-2 animate-bounce rounded-full bg-gray-400" />
+      <span className="h-2 w-2 animate-bounce rounded-full bg-gray-400" style={{ animationDelay: '0.1s' }} />
+      <span className="h-2 w-2 animate-bounce rounded-full bg-gray-400" style={{ animationDelay: '0.2s' }} />
+    </div>
+  )
+}
+
+export function ErrorBox({ msg }: { msg?: string }) {
+  return (
+    <div className="rounded-lg border border-red-200 bg-red-50 px-3 py-2 text-sm text-red-700">
+      {msg || 'Verbindung fehlgeschlagen'}
+    </div>
+  )
+}
+
+/** Answer mode body (stacked): summary + answer (with [n] coupling) + evidence/visual/footnotes. */
+export function AnswerBody({ response }: { response: AdvisorResponse }) {
+  const { highlightedId, cite } = useCitationHighlight(response.citations)
+  return (
+    <div className="space-y-3">
+      <EvidenceSummary response={response} />
+      <div className="rounded-lg border border-gray-200 bg-white px-3 py-2">
+        <Markdown content={response.answer || ''} citations={cite} />
+      </div>
+      <EvidencePane evidence={response.evidence} highlightedId={highlightedId} />
+      <VisualEvidencePane items={response.visual_evidence} />
+      <FootnotesPane footnotes={response.footnotes} />
+    </div>
+  )
+}
+
+/** One case rendered stacked (narrow mode). Clarify -> L1 + chips; answer -> full evidence body. */
+export function CaseView({
+  c,
+  onSelectContext,
+  busy,
+  showQuestion,
+  onRemove,
+}: {
+  c: AdvisorCase
+  onSelectContext: (ctx: string) => void
+  busy: boolean
+  showQuestion?: boolean
+  onRemove?: () => void
+}) {
+  const r = c.response
+  const { copiedKey, copy } = useClipboard()
+  return (
+    <div className="group space-y-2 border-b border-gray-100 pb-4 last:border-0">
+      <div className="flex items-start justify-between gap-2">
+        {showQuestion ? (
+          <div className="min-w-0 flex-1 text-xs text-gray-500">
+            <span className="font-medium text-gray-400">Frage:</span> {c.question}
+          </div>
+        ) : (
+          <span className="flex-1" />
+        )}
+        <div className="flex shrink-0 items-center gap-0.5 text-gray-400 opacity-0 transition-opacity group-hover:opacity-100">
+          <button
+            type="button"
+            title="Frage & Antwort kopieren"
+            aria-label="Frage & Antwort kopieren"
+            onClick={() => copy(c.id, formatCaseForCopy(c))}
+            className="rounded p-0.5 hover:bg-gray-100 hover:text-gray-700"
+          >
+            {copiedKey === c.id ? <Check className="h-3.5 w-3.5 text-green-600" /> : <Copy className="h-3.5 w-3.5" />}
+          </button>
+          {onRemove && (
+            <button
+              type="button"
+              title="Frage löschen"
+              aria-label="Frage löschen"
+              onClick={onRemove}
+              className="rounded p-0.5 hover:bg-gray-100 hover:text-gray-700"
+            >
+              <Trash2 className="h-3.5 w-3.5" />
+            </button>
+          )}
+        </div>
+      </div>
+      {c.status === 'loading' && <LoadingDots />}
+      {c.status === 'error' && <ErrorBox msg={c.error} />}
+      {r && r.mode === 'clarify' && (
+        <ClarifyView response={r} onSelectContext={onSelectContext} busy={busy} />
+      )}
+      {r && r.mode === 'answer' && <AnswerBody response={r} />}
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/ClarifyView.tsx

@@ -0,0 +1,52 @@
+'use client'
+
+import { Info } from 'lucide-react'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
+import { Markdown } from './Markdown'
+
+/**
+ * Clarify mode: a short general (L1) definition — explicitly marked as general, no legal source —
+ * plus domain context chips. Picking a chip re-runs the case scoped to that domain (-> L2).
+ */
+export function ClarifyView({
+  response,
+  onSelectContext,
+  busy,
+}: {
+  response: AdvisorResponse
+  onSelectContext: (id: string) => void
+  busy: boolean
+}) {
+  const chips = response.clarity.suggested_contexts ?? []
+  return (
+    <div className="space-y-3">
+      <div className="rounded-lg border border-amber-200 bg-amber-50 px-3 py-2">
+        <div className="mb-1 flex items-center gap-1 text-[11px] font-semibold text-amber-700">
+          <Info className="h-3.5 w-3.5" />
+          Allgemeine Definition (ohne Rechtsquelle)
+        </div>
+        <Markdown content={response.general_answer || ''} />
+      </div>
+      {chips.length > 0 && (
+        <div>
+          <div className="mb-1.5 text-xs font-medium text-gray-700">
+            Meintest du einen bestimmten Kontext?
+          </div>
+          <div className="flex flex-wrap gap-1.5">
+            {chips.map((c) => (
+              <button
+                key={c.id}
+                type="button"
+                disabled={busy}
+                onClick={() => onSelectContext(c.id)}
+                className="rounded-full border border-indigo-200 bg-white px-3 py-1 text-xs font-medium text-indigo-700 transition-colors hover:bg-indigo-50 disabled:opacity-50"
+              >
+                {c.label}
+              </button>
+            ))}
+          </div>
+        </div>
+      )}
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/EmptyState.tsx

@@ -0,0 +1,64 @@
+'use client'
+
+import { ShieldCheck } from 'lucide-react'
+
+export const EXAMPLE_QUESTIONS: Record<string, string[]> = {
+  vvt: [
+    'Was ist ein Verarbeitungsverzeichnis?',
+    'Welche Informationen muss ich erfassen?',
+    'Wie dokumentiere ich die Rechtsgrundlage?',
+  ],
+  'compliance-scope': [
+    'Was bedeutet L3?',
+    'Wann brauche ich eine DSFA?',
+    'Was ist der Unterschied zwischen L2 und L3?',
+  ],
+  tom: [
+    'Was sind TOM?',
+    'Welche Massnahmen sind erforderlich?',
+    'Wie dokumentiere ich Verschluesselung?',
+  ],
+  dsfa: ['Was ist eine DSFA?', 'Wann ist eine DSFA verpflichtend?', 'Wie bewerte ich Risiken?'],
+  loeschfristen: [
+    'Wie definiere ich Loeschfristen?',
+    'Unterschied Loeschpflicht und Aufbewahrungspflicht?',
+    'Wann muss ich Daten loeschen?',
+  ],
+  default: [
+    'Wie starte ich mit dem SDK?',
+    'Was ist der erste Schritt?',
+    'Welche Compliance-Anforderungen gelten fuer KI-Systeme?',
+  ],
+}
+
+export function AdvisorEmptyState({
+  exampleQuestions,
+  onExampleClick,
+}: {
+  exampleQuestions: string[]
+  onExampleClick: (question: string) => void
+}) {
+  return (
+    <div className="px-4 py-8 text-center">
+      <div className="mx-auto mb-3 flex h-14 w-14 items-center justify-center rounded-full bg-indigo-100">
+        <ShieldCheck className="h-7 w-7 text-indigo-600" />
+      </div>
+      <h3 className="text-sm font-semibold text-gray-900">Compliance Advisor</h3>
+      <p className="mx-auto mt-1 max-w-xs text-xs text-gray-500">
+        Antworten mit nachvollziehbaren Quellen, Fundstellen und — wo vorhanden — Original-Abbildungen.
+      </p>
+      <div className="mt-4 space-y-2 text-left">
+        <p className="text-xs font-medium text-gray-700">Beispielfragen</p>
+        {exampleQuestions.map((q, i) => (
+          <button
+            key={i}
+            onClick={() => onExampleClick(q)}
+            className="w-full rounded-lg border border-gray-200 bg-white px-3 py-2 text-left text-xs text-gray-700 transition-colors hover:bg-indigo-50"
+          >
+            {q}
+          </button>
+        ))}
+      </div>
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/EvidencePane.tsx

@@ -0,0 +1,76 @@
+'use client'
+
+import { useState } from 'react'
+import { ChevronDown, ChevronRight, Library } from 'lucide-react'
+import type { EvidenceUnit } from '@/lib/sdk/advisor/contract'
+import { resolveRegulation } from '@/lib/sdk/advisor/regulation-display'
+import { EvidenceUnitCard } from './EvidenceUnitCard'
+import { PaneHeader } from './PaneHeader'
+
+interface Group {
+  key: string
+  label: string
+  units: EvidenceUnit[]
+}
+
+function groupByFamily(units: EvidenceUnit[]): Group[] {
+  const map = new Map<string, Group>()
+  for (const u of units) {
+    const d = resolveRegulation({ code: u.document, short: u.document })
+    const g = map.get(d.familyKey) ?? { key: d.familyKey, label: d.familyLabel, units: [] }
+    g.units.push(u)
+    map.set(d.familyKey, g)
+  }
+  return [...map.values()].sort((a, b) => b.units.length - a.units.length)
+}
+
+function EvidenceGroup({ group, highlightedId }: { group: Group; highlightedId?: string }) {
+  const [open, setOpen] = useState(group.units.length <= 3)
+  return (
+    <div className="rounded-lg border border-gray-200 bg-white">
+      <button
+        type="button"
+        onClick={() => setOpen((v) => !v)}
+        className="flex w-full items-center justify-between gap-2 px-2.5 py-2 text-left"
+      >
+        <span className="min-w-0 truncate text-xs font-semibold text-gray-900">{group.label}</span>
+        <span className="flex flex-shrink-0 items-center gap-1 text-[11px] text-gray-500">
+          {group.units.length} Treffer
+          {open ? <ChevronDown className="h-3.5 w-3.5" /> : <ChevronRight className="h-3.5 w-3.5" />}
+        </span>
+      </button>
+      {open && (
+        <div className="space-y-1 border-t border-gray-100 px-2 py-2">
+          {group.units.map((u) => (
+            <EvidenceUnitCard key={u.evidence_id} unit={u} compact highlighted={u.evidence_id === highlightedId} />
+          ))}
+        </div>
+      )}
+    </div>
+  )
+}
+
+/** Evidence pane — units grouped by document/regulation family, count + expandable. */
+export function EvidencePane({
+  evidence,
+  highlightedId,
+}: {
+  evidence: EvidenceUnit[]
+  highlightedId?: string
+}) {
+  const groups = groupByFamily(evidence)
+  return (
+    <section>
+      <PaneHeader icon={<Library className="h-3.5 w-3.5 text-gray-500" />} title="Evidence" count={evidence.length} />
+      {groups.length === 0 ? (
+        <p className="px-1 text-[11px] text-gray-400">Keine strukturierte Evidence zu dieser Antwort.</p>
+      ) : (
+        <div className="space-y-1.5">
+          {groups.map((g) => (
+            <EvidenceGroup key={g.key} group={g} highlightedId={highlightedId} />
+          ))}
+        </div>
+      )}
+    </section>
+  )
+}

admin-compliance/components/sdk/advisor/EvidenceSummary.tsx

@@ -0,0 +1,135 @@
+'use client'
+
+import { BookMarked, FileText, Hash, Image as ImageIcon, Library, Scale } from 'lucide-react'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
+import {
+  provisionSummary,
+  summarizeEvidence,
+  type FamilyGroup,
+} from '@/lib/sdk/advisor/evidence-grouping'
+
+const plural = (n: number, one: string, many: string) => (n === 1 ? one : many)
+
+function Count({
+  icon,
+  value,
+  label,
+  dim,
+}: {
+  icon: React.ReactNode
+  value: number
+  label: string
+  dim?: boolean
+}) {
+  return (
+    <div
+      className={`flex items-center gap-2 rounded-lg border px-2.5 py-1.5 ${
+        dim ? 'border-gray-100 bg-gray-50' : 'border-gray-200 bg-white'
+      }`}
+    >
+      <span className={dim ? 'text-gray-300' : 'text-indigo-500'}>{icon}</span>
+      <span>
+        <span className={`text-sm font-bold ${dim ? 'text-gray-400' : 'text-gray-900'}`}>{value}</span>{' '}
+        <span className="text-[11px] text-gray-500">{label}</span>
+      </span>
+    </div>
+  )
+}
+
+function GroupRow({ group, icon }: { group: FamilyGroup; icon: React.ReactNode }) {
+  // A single-unit guidance doc needs no "1 Fundstelle" noise; norms always show their provisions.
+  const detail = group.sections.length === 0 && group.units <= 1 ? '' : provisionSummary(group)
+  return (
+    <div className="flex items-start gap-2 py-0.5">
+      <span className="mt-0.5 shrink-0 text-gray-400">{icon}</span>
+      <span className="min-w-0 flex-1 text-[12px] leading-snug text-gray-700">{group.label}</span>
+      {detail && <span className="shrink-0 text-[11px] font-medium text-gray-500">{detail}</span>}
+    </div>
+  )
+}
+
+function Section({
+  title,
+  groups,
+  icon,
+}: {
+  title: string
+  groups: FamilyGroup[]
+  icon: React.ReactNode
+}) {
+  if (groups.length === 0) return null
+  return (
+    <div className="mt-2 first:mt-0">
+      <div className="mb-0.5 text-[10px] font-semibold uppercase tracking-wide text-gray-400">{title}</div>
+      {groups.map((g) => (
+        <GroupRow key={g.key} group={g} icon={icon} />
+      ))}
+    </div>
+  )
+}
+
+/**
+ * "Diese Antwort stützt sich auf" — describes the EVIDENCE (not the documents), objective counts
+ * only (no fabricated trust score). When the Legal-KG ships `bindingness`, binding Rechtsgrundlagen
+ * are split from Leitlinien (soft-law guidance); until then it shows a neutral evidence breakdown.
+ */
+export function EvidenceSummary({ response }: { response: AdvisorResponse }) {
+  const m = summarizeEvidence(response.evidence)
+  const figures = response.visual_evidence.length
+  const notes = response.footnotes.length
+  const cls = 'h-4 w-4'
+  const smallIcon = 'h-3.5 w-3.5'
+
+  return (
+    <div>
+      <div className="mb-1.5 text-[10px] font-semibold uppercase tracking-wide text-gray-400">
+        Diese Antwort stützt sich auf
+      </div>
+
+      <div className="grid grid-cols-2 gap-1.5">
+        {m.hasBindingness && (
+          <>
+            <Count
+              icon={<Scale className={cls} />}
+              value={m.normProvisions}
+              label={plural(m.normProvisions, 'Rechtsgrundlage', 'Rechtsgrundlagen')}
+            />
+            <Count
+              icon={<BookMarked className={cls} />}
+              value={m.guidanceCount}
+              label={plural(m.guidanceCount, 'Leitlinie', 'Leitlinien')}
+              dim={m.guidanceCount === 0}
+            />
+          </>
+        )}
+        <Count
+          icon={<ImageIcon className={cls} />}
+          value={figures}
+          label={plural(figures, 'Abbildung', 'Abbildungen')}
+          dim={figures === 0}
+        />
+        <Count
+          icon={<Hash className={cls} />}
+          value={notes}
+          label={plural(notes, 'Fußnote', 'Fußnoten')}
+          dim={notes === 0}
+        />
+        <Count icon={<FileText className={cls} />} value={m.unitCount} label="Evidence Units" />
+      </div>
+
+      {m.groups.length > 0 && (
+        <div className="mt-2.5 rounded-lg border border-gray-100 bg-gray-50/60 px-2.5 py-1.5">
+          {m.hasBindingness ? (
+            <>
+              <Section title="Rechtsgrundlagen" groups={m.norms} icon={<Scale className={smallIcon} />} />
+              <Section title="Leitlinien" groups={m.guidance} icon={<BookMarked className={smallIcon} />} />
+              <Section title="Weitere" groups={m.other} icon={<FileText className={smallIcon} />} />
+            </>
+          ) : (
+            m.groups.map((g) => <GroupRow key={g.key} group={g} icon={<Library className={smallIcon} />} />)
+          )}
+        </div>
+      )}
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/EvidenceUnitCard.tsx

@@ -0,0 +1,80 @@
+'use client'
+
+import { useState } from 'react'
+import { ChevronDown, ChevronRight, ExternalLink } from 'lucide-react'
+import type { EvidenceUnit } from '@/lib/sdk/advisor/contract'
+import { resolveRegulation } from '@/lib/sdk/advisor/regulation-display'
+
+/** One evidence unit (contract shape). Compact inside a document group: chapter/section only. */
+export function EvidenceUnitCard({
+  unit,
+  compact,
+  highlighted,
+}: {
+  unit: EvidenceUnit
+  compact?: boolean
+  highlighted?: boolean
+}) {
+  const [open, setOpen] = useState(false)
+  const d = resolveRegulation({ code: unit.document, short: unit.document })
+  const crumbs = [unit.section, unit.paragraph].filter((x): x is string => Boolean(x))
+  const canOpen = !!unit.url && /^https?:\/\//i.test(unit.url)
+
+  const header = compact ? (d.chapter ? `Kapitel ${d.chapter}` : crumbs[0] || d.familyLabel) : d.familyLabel
+  const sub = compact && !d.chapter && crumbs.length ? crumbs.slice(1) : crumbs
+
+  return (
+    <div
+      id={`ev-${unit.evidence_id}`}
+      className={`${
+        compact
+          ? 'rounded-md border border-gray-100 bg-gray-50 p-2'
+          : 'rounded-lg border border-gray-200 bg-white p-2.5'
+      } ${highlighted ? 'ring-2 ring-indigo-400' : ''}`}
+    >
+      <div className="flex items-start justify-between gap-2">
+        <div className="min-w-0">
+          <div className="truncate text-xs font-semibold text-gray-900">{header}</div>
+          {sub.length > 0 && (
+            <div className="mt-0.5 flex flex-wrap items-center gap-x-1 text-[11px] text-gray-500">
+              {sub.map((c, i) => (
+                <span key={i} className="flex items-center gap-1">
+                  {i > 0 && <span className="text-gray-300">›</span>}
+                  {c}
+                </span>
+              ))}
+            </div>
+          )}
+        </div>
+        {canOpen && (
+          <a
+            href={unit.url}
+            target="_blank"
+            rel="noopener noreferrer"
+            className="flex flex-shrink-0 items-center gap-0.5 rounded px-1.5 py-0.5 text-[11px] font-medium text-indigo-600 hover:bg-indigo-50"
+          >
+            <ExternalLink className="h-3 w-3" />
+            öffnen
+          </a>
+        )}
+      </div>
+      {unit.snippet && (
+        <div className="mt-1.5">
+          <button
+            type="button"
+            onClick={() => setOpen((v) => !v)}
+            className="flex items-center gap-0.5 text-[11px] text-gray-400 hover:text-gray-600"
+          >
+            {open ? <ChevronDown className="h-3 w-3" /> : <ChevronRight className="h-3 w-3" />}
+            Textauszug
+          </button>
+          {open && (
+            <p className="mt-1 border-l-2 border-gray-200 pl-2 text-[11px] italic text-gray-500">
+              {unit.snippet}
+            </p>
+          )}
+        </div>
+      )}
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/EvidenceWorkspace.tsx

@@ -0,0 +1,134 @@
+'use client'
+
+import { useEffect, useRef } from 'react'
+import type { AdvisorCase, AdvisorThread } from './useAdvisorCase'
+import { StickyQuestion } from './StickyQuestion'
+import { AdvisorEmptyState } from './EmptyState'
+import { CaseView, LoadingDots, ErrorBox } from './CaseView'
+import { ClarifyView } from './ClarifyView'
+import { EvidenceSummary } from './EvidenceSummary'
+import { EvidencePane } from './EvidencePane'
+import { VisualEvidencePane } from './VisualEvidencePane'
+import { FootnotesPane } from './FootnotesPane'
+import { Markdown } from './Markdown'
+import { ThreadMenu } from './ThreadMenu'
+import { useCitationHighlight } from './useCitationHighlight'
+
+/**
+ * Advisor body as topic THREADS of cases.
+ * - Narrow: stacked cases with a pinned last question; per-case copy + delete.
+ * - Wide/fullscreen: 3-column Case Workspace — topic tree (left) | answer/clarify (center) | evidence (right).
+ */
+export function EvidenceWorkspace({
+  cases,
+  threads,
+  expanded,
+  busy,
+  activeCaseId,
+  exampleQuestions,
+  onExample,
+  onSelectContext,
+  onSelectCase,
+  onRemove,
+}: {
+  cases: AdvisorCase[]
+  threads: AdvisorThread[]
+  expanded: boolean
+  busy: boolean
+  activeCaseId: string | null
+  exampleQuestions: string[]
+  onExample: (q: string) => void
+  onSelectContext: (caseId: string, ctx: string) => void
+  onSelectCase: (id: string) => void
+  onRemove: (id: string) => void
+}) {
+  const endRef = useRef<HTMLDivElement>(null)
+  const latest = cases[cases.length - 1]
+  const active = cases.find((c) => c.id === activeCaseId) ?? latest
+
+  useEffect(() => {
+    if (!expanded) endRef.current?.scrollIntoView({ behavior: 'smooth' })
+  }, [cases.length, expanded])
+
+  const answer = active?.response?.mode === 'answer' ? active.response : null
+  const { highlightedId, cite } = useCitationHighlight(answer?.citations ?? [])
+
+  if (cases.length === 0) {
+    return (
+      <div className="flex-1 overflow-y-auto bg-gray-50">
+        <AdvisorEmptyState exampleQuestions={exampleQuestions} onExampleClick={onExample} />
+      </div>
+    )
+  }
+
+  if (!expanded) {
+    return (
+      <div className="min-h-0 flex-1 overflow-y-auto bg-gray-50">
+        {latest && <StickyQuestion question={latest.question} />}
+        <div className="space-y-4 p-4">
+          {cases.map((c, i) => (
+            <CaseView
+              key={c.id}
+              c={c}
+              busy={busy}
+              showQuestion={i !== cases.length - 1}
+              onSelectContext={(ctx) => onSelectContext(c.id, ctx)}
+              onRemove={() => onRemove(c.id)}
+            />
+          ))}
+          <div ref={endRef} />
+        </div>
+      </div>
+    )
+  }
+
+  const r = active?.response
+  return (
+    <div className="grid min-h-0 flex-1 grid-cols-[260px_1fr_320px] divide-x divide-gray-200 overflow-hidden">
+      <aside className="min-h-0 overflow-y-auto bg-indigo-50/40 p-3">
+        <ThreadMenu
+          threads={threads}
+          activeCaseId={active?.id ?? null}
+          onSelectCase={onSelectCase}
+          onRemove={onRemove}
+        />
+        {active && (
+          <div className="mt-3 border-t border-indigo-100 pt-3">
+            <div className="text-[10px] font-semibold uppercase tracking-wide text-indigo-400">Aktive Frage</div>
+            <div className="mb-3 text-sm font-medium text-gray-800">{active.question}</div>
+            {answer && <EvidenceSummary response={answer} />}
+          </div>
+        )}
+      </aside>
+
+      <main className="min-h-0 overflow-y-auto bg-gray-50 p-4">
+        {active?.status === 'loading' && <LoadingDots />}
+        {active?.status === 'error' && <ErrorBox msg={active.error} />}
+        {r?.mode === 'clarify' && (
+          <ClarifyView
+            response={r}
+            busy={busy}
+            onSelectContext={(ctx) => active && onSelectContext(active.id, ctx)}
+          />
+        )}
+        {r?.mode === 'answer' && (
+          <div className="rounded-lg border border-gray-200 bg-white px-3 py-2">
+            <Markdown content={r.answer || ''} citations={cite} />
+          </div>
+        )}
+      </main>
+
+      <aside className="min-h-0 space-y-3 overflow-y-auto bg-gray-50 p-3">
+        {answer ? (
+          <>
+            <EvidencePane evidence={answer.evidence} highlightedId={highlightedId} />
+            <VisualEvidencePane items={answer.visual_evidence} />
+            <FootnotesPane footnotes={answer.footnotes} />
+          </>
+        ) : (
+          <p className="px-1 text-[11px] text-gray-400">Evidence erscheint nach Auswahl eines Kontexts.</p>
+        )}
+      </aside>
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/FootnotesPane.tsx

@@ -0,0 +1,30 @@
+'use client'
+
+import { Hash } from 'lucide-react'
+import type { Footnote } from '@/lib/sdk/advisor/contract'
+import { PaneHeader } from './PaneHeader'
+
+/** Footnotes pane (C-FN) — rendered only when present. */
+export function FootnotesPane({ footnotes }: { footnotes: Footnote[] }) {
+  if (footnotes.length === 0) return null
+  return (
+    <section>
+      <PaneHeader icon={<Hash className="h-3.5 w-3.5 text-gray-500" />} title="Fußnoten" count={footnotes.length} />
+      <div className="space-y-1">
+        {footnotes.map((fn, i) => (
+          <div key={fn.footnote_id || i} className="rounded-md border border-gray-200 bg-white p-2 text-[11px]">
+            <span className="font-semibold text-gray-900">{fn.ref || `Fußnote ${i + 1}`}</span>
+            {(fn.document || fn.section) && (
+              <span className="text-gray-400">
+                {' · '}
+                {fn.document}
+                {fn.section ? ` / ${fn.section}` : ''}
+              </span>
+            )}
+            {fn.text && <p className="mt-0.5 text-gray-600">{fn.text}</p>}
+          </div>
+        ))}
+      </div>
+    </section>
+  )
+}

admin-compliance/components/sdk/advisor/Markdown.test.tsx

@@ -0,0 +1,40 @@
+import { describe, it, expect } from 'vitest'
+import { render } from '@testing-library/react'
+import { Markdown } from './Markdown'
+
+describe('Markdown', () => {
+  it('renders headings, bold and bullet lists (not raw markdown markers)', () => {
+    const { container } = render(
+      <Markdown
+        content={'## Pflichten\n\nDer **Verantwortliche** muss:\n\n- ein Verzeichnis fuehren\n- Risiken bewerten'}
+      />,
+    )
+    expect(container.querySelector('h4')?.textContent).toBe('Pflichten')
+    expect(container.querySelector('strong')?.textContent).toBe('Verantwortliche')
+    expect(container.querySelectorAll('li')).toHaveLength(2)
+    expect(container.textContent).not.toContain('##')
+    expect(container.textContent).not.toContain('**')
+  })
+
+  it('renders ordered lists and inline code', () => {
+    const { container } = render(<Markdown content={'1. Erst `init`\n2. Dann `build`'} />)
+    expect(container.querySelector('ol')).not.toBeNull()
+    expect(container.querySelectorAll('li')).toHaveLength(2)
+    expect(container.querySelectorAll('code')).toHaveLength(2)
+  })
+
+  it('renders fenced code blocks', () => {
+    const { container } = render(<Markdown content={'```\nconst x = 1\n```'} />)
+    expect(container.querySelector('pre')).not.toBeNull()
+    expect(container.textContent).toContain('const x = 1')
+  })
+
+  it('only allows http(s) links', () => {
+    const { container } = render(
+      <Markdown content={'[ok](https://example.test) and [bad](javascript:alert(1))'} />,
+    )
+    const links = container.querySelectorAll('a')
+    expect(links).toHaveLength(1)
+    expect(links[0].getAttribute('href')).toBe('https://example.test')
+  })
+})

admin-compliance/components/sdk/advisor/Markdown.tsx

@@ -0,0 +1,176 @@
+'use client'
+
+// Minimal, SAFE markdown -> React renderer. No dangerouslySetInnerHTML, no dependency.
+// Covers the subset LLMs emit: headings, bold, italic, inline code, fenced code, ul/ol, links.
+// Plus deliberate [n] citation markers (mapped via `citations`, NOT parsed for structure).
+
+export interface CiteHandler {
+  count: number
+  onSelect: (n: number) => void
+}
+
+const INLINE_RE =
+  /(`[^`]+`|\*\*[^*]+\*\*|\*[^*\s][^*]*\*|_[^_]+_|\[[^\]]+\]\([^)]+\)|\[\d+\])/g
+
+function renderInline(text: string, kp: string, cite?: CiteHandler): React.ReactNode[] {
+  const nodes: React.ReactNode[] = []
+  let last = 0
+  let idx = 0
+  INLINE_RE.lastIndex = 0
+  let m: RegExpExecArray | null
+  while ((m = INLINE_RE.exec(text)) !== null) {
+    if (m.index > last) nodes.push(text.slice(last, m.index))
+    const tok = m[0]
+    const key = `${kp}-${idx++}`
+    if (tok.startsWith('`')) {
+      nodes.push(
+        <code key={key} className="rounded bg-gray-100 px-1 py-0.5 font-mono text-[0.85em]">
+          {tok.slice(1, -1)}
+        </code>,
+      )
+    } else if (tok.startsWith('**')) {
+      nodes.push(
+        <strong key={key} className="font-semibold text-gray-900">
+          {tok.slice(2, -2)}
+        </strong>,
+      )
+    } else if (tok.startsWith('*') || tok.startsWith('_')) {
+      nodes.push(<em key={key}>{tok.slice(1, -1)}</em>)
+    } else if (/^\[\d+\]$/.test(tok)) {
+      const n = parseInt(tok.slice(1, -1), 10)
+      if (cite && n >= 1 && n <= cite.count) {
+        nodes.push(
+          <button
+            key={key}
+            type="button"
+            onClick={() => cite.onSelect(n)}
+            className="mx-0.5 align-super text-[10px] font-semibold text-indigo-600 hover:underline"
+            title={`Beleg ${n} anzeigen`}
+          >
+            [{n}]
+          </button>,
+        )
+      } else {
+        nodes.push(tok)
+      }
+    } else {
+      const mm = /^\[([^\]]+)\]\(([^)]+)\)$/.exec(tok)
+      if (mm && /^https?:\/\//i.test(mm[2])) {
+        nodes.push(
+          <a
+            key={key}
+            href={mm[2]}
+            target="_blank"
+            rel="noopener noreferrer"
+            className="text-indigo-600 underline hover:text-indigo-800"
+          >
+            {mm[1]}
+          </a>,
+        )
+      } else {
+        nodes.push(mm ? mm[1] : tok)
+      }
+    }
+    last = m.index + tok.length
+  }
+  if (last < text.length) nodes.push(text.slice(last))
+  return nodes
+}
+
+function Heading({ level, kp, text, cite }: { level: number; kp: string; text: string; cite?: CiteHandler }) {
+  const children = renderInline(text, kp, cite)
+  if (level <= 1) return <h3 className="mb-1 mt-3 text-base font-bold text-gray-900">{children}</h3>
+  if (level === 2) return <h4 className="mb-1 mt-3 text-sm font-bold text-gray-900">{children}</h4>
+  return <h5 className="mb-1 mt-2 text-sm font-semibold text-gray-800">{children}</h5>
+}
+
+const UL_RE = /^\s*[-*]\s+/
+const OL_RE = /^\s*\d+\.\s+/
+const H_RE = /^(#{1,6})\s+(.*)$/
+
+export function Markdown({ content, citations }: { content: string; citations?: CiteHandler }) {
+  const lines = (content || '').replace(/\r\n/g, '\n').split('\n')
+  const blocks: React.ReactNode[] = []
+  let i = 0
+  while (i < lines.length) {
+    const line = lines[i]
+    const key = `b${blocks.length}`
+
+    if (line.trim().startsWith('```')) {
+      const buf: string[] = []
+      i++
+      while (i < lines.length && !lines[i].trim().startsWith('```')) {
+        buf.push(lines[i])
+        i++
+      }
+      i++
+      blocks.push(
+        <pre
+          key={key}
+          className="my-2 overflow-x-auto rounded bg-gray-900 p-3 font-mono text-xs text-gray-100"
+        >
+          <code>{buf.join('\n')}</code>
+        </pre>,
+      )
+      continue
+    }
+    if (line.trim() === '') {
+      i++
+      continue
+    }
+    const h = H_RE.exec(line)
+    if (h) {
+      blocks.push(<Heading key={key} kp={key} level={h[1].length} text={h[2]} cite={citations} />)
+      i++
+      continue
+    }
+    if (UL_RE.test(line)) {
+      const items: string[] = []
+      while (i < lines.length && UL_RE.test(lines[i])) {
+        items.push(lines[i].replace(UL_RE, ''))
+        i++
+      }
+      blocks.push(
+        <ul key={key} className="my-1.5 ml-4 list-disc space-y-1 text-gray-700">
+          {items.map((it, k) => (
+            <li key={k}>{renderInline(it, `${key}-${k}`, citations)}</li>
+          ))}
+        </ul>,
+      )
+      continue
+    }
+    if (OL_RE.test(line)) {
+      const items: string[] = []
+      while (i < lines.length && OL_RE.test(lines[i])) {
+        items.push(lines[i].replace(OL_RE, ''))
+        i++
+      }
+      blocks.push(
+        <ol key={key} className="my-1.5 ml-5 list-decimal space-y-1 text-gray-700">
+          {items.map((it, k) => (
+            <li key={k}>{renderInline(it, `${key}-${k}`, citations)}</li>
+          ))}
+        </ol>,
+      )
+      continue
+    }
+    const para: string[] = []
+    while (
+      i < lines.length &&
+      lines[i].trim() !== '' &&
+      !H_RE.test(lines[i]) &&
+      !UL_RE.test(lines[i]) &&
+      !OL_RE.test(lines[i]) &&
+      !lines[i].trim().startsWith('```')
+    ) {
+      para.push(lines[i])
+      i++
+    }
+    blocks.push(
+      <p key={key} className="my-1.5 leading-relaxed text-gray-700">
+        {renderInline(para.join(' '), key, citations)}
+      </p>,
+    )
+  }
+  return <div className="advisor-markdown text-sm">{blocks}</div>
+}

admin-compliance/components/sdk/advisor/PaneHeader.tsx

@@ -0,0 +1,24 @@
+'use client'
+
+/** Shared section header for evidence panes (icon + title + count badge). */
+export function PaneHeader({
+  icon,
+  title,
+  count,
+}: {
+  icon: React.ReactNode
+  title: string
+  count?: number
+}) {
+  return (
+    <div className="mb-1.5 flex items-center gap-1.5 text-xs font-semibold text-gray-700">
+      {icon}
+      <span>{title}</span>
+      {count != null && (
+        <span className="rounded-full bg-gray-100 px-1.5 text-[10px] font-medium text-gray-500">
+          {count}
+        </span>
+      )}
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/StickyQuestion.tsx

@@ -0,0 +1,21 @@
+'use client'
+
+import { HelpCircle } from 'lucide-react'
+
+/** The last question, pinned so it never scrolls out of view while the answer grows. */
+export function StickyQuestion({ question }: { question: string }) {
+  if (!question) return null
+  return (
+    <div className="sticky top-0 z-10 border-b border-indigo-100 bg-indigo-50/95 px-4 py-2 backdrop-blur">
+      <div className="flex items-start gap-2">
+        <HelpCircle className="mt-0.5 h-4 w-4 flex-shrink-0 text-indigo-500" />
+        <div className="min-w-0">
+          <div className="text-[10px] font-semibold uppercase tracking-wide text-indigo-400">
+            Letzte Frage
+          </div>
+          <div className="text-sm font-medium text-gray-800">{question}</div>
+        </div>
+      </div>
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/ThreadMenu.tsx

@@ -0,0 +1,139 @@
+'use client'
+
+import { useState } from 'react'
+import { Check, ChevronDown, ChevronRight, Copy, Files, Trash2 } from 'lucide-react'
+import type { AdvisorThread } from './useAdvisorCase'
+import { formatCaseForCopy, formatThreadForCopy } from '@/lib/sdk/advisor/copy'
+import { useClipboard } from './useClipboard'
+
+function IconBtn({
+  title,
+  onClick,
+  children,
+}: {
+  title: string
+  onClick: () => void
+  children: React.ReactNode
+}) {
+  return (
+    <button
+      type="button"
+      title={title}
+      aria-label={title}
+      onClick={(e) => {
+        e.stopPropagation()
+        onClick()
+      }}
+      className="rounded p-0.5 text-gray-400 transition-colors hover:bg-gray-200 hover:text-gray-700"
+    >
+      {children}
+    </button>
+  )
+}
+
+/**
+ * Left-menu topic tree: each thread's first question is the Thema; follow-ups nest underneath and
+ * expand/collapse. Per row: copy (single Q&A) + delete; per topic: copy the whole thread.
+ */
+export function ThreadMenu({
+  threads,
+  activeCaseId,
+  onSelectCase,
+  onRemove,
+}: {
+  threads: AdvisorThread[]
+  activeCaseId: string | null
+  onSelectCase: (id: string) => void
+  onRemove: (id: string) => void
+}) {
+  const { copiedKey, copy } = useClipboard()
+  const [collapsed, setCollapsed] = useState<Record<string, boolean>>({})
+  const ic = 'h-3.5 w-3.5'
+
+  return (
+    <div>
+      <div className="mb-1 text-[10px] font-semibold uppercase tracking-wide text-gray-400">Themen</div>
+      <div className="space-y-0.5">
+        {threads.map((t) => {
+          const first = t.cases[0]
+          const followups = t.cases.slice(1)
+          const open = !collapsed[t.id]
+          const activeInThread = t.cases.some((c) => c.id === activeCaseId)
+          return (
+            <div key={t.id}>
+              <div
+                className={`group flex items-center gap-1 rounded px-1.5 py-1 ${
+                  first.id === activeCaseId ? 'bg-indigo-100' : activeInThread ? 'bg-indigo-50/60' : 'hover:bg-gray-100'
+                }`}
+              >
+                {followups.length > 0 ? (
+                  <button
+                    type="button"
+                    onClick={() => setCollapsed((s) => ({ ...s, [t.id]: !s[t.id] }))}
+                    className="text-gray-400 hover:text-gray-700"
+                    aria-label={open ? 'Thema einklappen' : 'Thema aufklappen'}
+                  >
+                    {open ? <ChevronDown className={ic} /> : <ChevronRight className={ic} />}
+                  </button>
+                ) : (
+                  <span className="w-3.5 shrink-0" />
+                )}
+                <button
+                  type="button"
+                  onClick={() => onSelectCase(first.id)}
+                  title={first.question}
+                  className={`min-w-0 flex-1 truncate text-left text-[12px] font-medium ${
+                    first.id === activeCaseId ? 'text-indigo-800' : 'text-gray-700'
+                  }`}
+                >
+                  {first.question}
+                </button>
+                <div className="flex shrink-0 items-center gap-0.5 opacity-0 transition-opacity group-hover:opacity-100">
+                  <IconBtn title="Ganzes Thema kopieren" onClick={() => copy(`thread:${t.id}`, formatThreadForCopy(t.title, t.cases))}>
+                    {copiedKey === `thread:${t.id}` ? <Check className={`${ic} text-green-600`} /> : <Files className={ic} />}
+                  </IconBtn>
+                  <IconBtn title="Diese Frage kopieren" onClick={() => copy(`case:${first.id}`, formatCaseForCopy(first))}>
+                    {copiedKey === `case:${first.id}` ? <Check className={`${ic} text-green-600`} /> : <Copy className={ic} />}
+                  </IconBtn>
+                  <IconBtn title="Frage löschen" onClick={() => onRemove(first.id)}>
+                    <Trash2 className={ic} />
+                  </IconBtn>
+                </div>
+              </div>
+
+              {open &&
+                followups.map((c) => (
+                  <div
+                    key={c.id}
+                    className={`group ml-4 flex items-center gap-1 rounded px-1.5 py-1 ${
+                      c.id === activeCaseId ? 'bg-indigo-100' : 'hover:bg-gray-100'
+                    }`}
+                  >
+                    <span className="shrink-0 text-gray-300">•</span>
+                    <button
+                      type="button"
+                      onClick={() => onSelectCase(c.id)}
+                      title={c.question}
+                      className={`min-w-0 flex-1 truncate text-left text-[11px] ${
+                        c.id === activeCaseId ? 'text-indigo-800' : 'text-gray-600'
+                      }`}
+                    >
+                      {c.question}
+                    </button>
+                    <div className="flex shrink-0 items-center gap-0.5 opacity-0 transition-opacity group-hover:opacity-100">
+                      <IconBtn title="Diese Frage kopieren" onClick={() => copy(`case:${c.id}`, formatCaseForCopy(c))}>
+                        {copiedKey === `case:${c.id}` ? <Check className={`${ic} text-green-600`} /> : <Copy className={ic} />}
+                      </IconBtn>
+                      <IconBtn title="Frage löschen" onClick={() => onRemove(c.id)}>
+                        <Trash2 className={ic} />
+                      </IconBtn>
+                    </div>
+                  </div>
+                ))}
+            </div>
+          )
+        })}
+      </div>
+    </div>
+  )
+}

admin-compliance/components/sdk/advisor/VisualEvidencePane.tsx

@@ -0,0 +1,70 @@
+'use client'
+
+import { ExternalLink, Image as ImageIcon } from 'lucide-react'
+import type { VisualEvidence } from '@/lib/sdk/advisor/contract'
+import { PaneHeader } from './PaneHeader'
+
+function VisualCard({ v }: { v: VisualEvidence }) {
+  const canOpen = !!v.image_ref && /^https?:\/\//i.test(v.image_ref)
+  return (
+    <div className="rounded-lg border border-gray-200 bg-white p-2.5">
+      <div className="flex items-start justify-between gap-2">
+        <div className="min-w-0">
+          <div className="text-xs font-semibold text-gray-900">{v.caption || v.visual_type}</div>
+          <div className="mt-0.5 flex flex-wrap items-center gap-1 text-[11px] text-gray-500">
+            <span className="rounded bg-gray-100 px-1 text-[10px] uppercase tracking-wide text-gray-500">
+              {v.visual_type}
+            </span>
+            <span>Quelle: {v.document}</span>
+          </div>
+        </div>
+        {canOpen && (
+          <a
+            href={v.image_ref}
+            target="_blank"
+            rel="noopener noreferrer"
+            className="flex flex-shrink-0 items-center gap-0.5 rounded px-1.5 py-0.5 text-[11px] font-medium text-indigo-600 hover:bg-indigo-50"
+          >
+            <ExternalLink className="h-3 w-3" />
+            Original anzeigen
+          </a>
+        )}
+      </div>
+      {canOpen ? (
+        <a href={v.image_ref} target="_blank" rel="noopener noreferrer" className="mt-1.5 block">
+          {/* eslint-disable-next-line @next/next/no-img-element */}
+          <img
+            src={v.image_ref}
+            alt={v.caption || v.visual_type}
+            loading="lazy"
+            className="max-h-44 w-full rounded border border-gray-100 object-contain"
+          />
+        </a>
+      ) : (
+        <div className="mt-1.5 flex items-center justify-center rounded border border-dashed border-gray-200 bg-gray-50 px-3 py-5 text-[11px] text-gray-400">
+          Original-Darstellung folgt
+        </div>
+      )}
+      {v.vision_summary && <p className="mt-1.5 text-[11px] italic text-gray-500">{v.vision_summary}</p>}
+    </div>
+  )
+}
+
+/** Visual evidence (C8) — diagrams/figures, rendered only when present. */
+export function VisualEvidencePane({ items }: { items: VisualEvidence[] }) {
+  if (items.length === 0) return null
+  return (
+    <section>
+      <PaneHeader
+        icon={<ImageIcon className="h-3.5 w-3.5 text-gray-500" />}
+        title="Diagramme & Abbildungen"
+        count={items.length}
+      />
+      <div className="space-y-1.5">
+        {items.map((v) => (
+          <VisualCard key={v.visual_id} v={v} />
+        ))}
+      </div>
+    </section>
+  )
+}

admin-compliance/components/sdk/advisor/useAdvisorCase.ts

@@ -0,0 +1,188 @@
+'use client'
+
+import { useCallback, useEffect, useMemo, useRef, useState } from 'react'
+import type { AdvisorResponse } from '@/lib/sdk/advisor/contract'
+
+export interface AdvisorCase {
+  id: string
+  threadId: string
+  question: string
+  response: AdvisorResponse | null
+  selectedContext: string | null
+  status: 'loading' | 'done' | 'error'
+  error?: string
+}
+
+/** A topic: the first case's question is the title; follow-ups are the rest, in order. */
+export interface AdvisorThread {
+  id: string
+  title: string
+  cases: AdvisorCase[]
+}
+
+interface HistoryTurn {
+  role: 'user' | 'assistant'
+  content: string
+}
+
+interface UseAdvisorCaseArgs {
+  currentStep: string
+  country: string
+}
+
+let counter = 0
+const uid = (p: string) => `${p}-${Date.now()}-${counter++}`
+
+/**
+ * Drives the Advisor as topic THREADS of CASES. Each ask posts {question, context?, history} and
+ * receives a structured AdvisorResponse (clarify | answer) — no streaming, no answer-text parsing.
+ * A follow-up appends to the active thread (and carries the thread's prior Q&A as history);
+ * newTopic() starts a fresh thread. selectContext() re-runs a case scoped to a chosen domain.
+ */
+export function useAdvisorCase({ currentStep, country }: UseAdvisorCaseArgs) {
+  const [cases, setCases] = useState<AdvisorCase[]>([])
+  const [busy, setBusy] = useState(false)
+  const [activeCaseId, setActiveCaseId] = useState<string | null>(null)
+  const [activeThreadId, setActiveThreadId] = useState<string | null>(null)
+  const abortRef = useRef<AbortController | null>(null)
+
+  const patch = useCallback((id: string, p: Partial<AdvisorCase>) => {
+    setCases((prev) => prev.map((c) => (c.id === id ? { ...c, ...p } : c)))
+  }, [])
+
+  // Prior answered turns of a thread, up to (but excluding) `beforeId`, for contextual follow-ups.
+  const buildHistory = useCallback(
+    (threadId: string, beforeId?: string): HistoryTurn[] => {
+      const turns: HistoryTurn[] = []
+      for (const c of cases) {
+        if (c.threadId !== threadId) continue
+        if (beforeId && c.id === beforeId) break
+        const a = c.response?.answer ?? c.response?.general_answer
+        if (!a) continue
+        turns.push({ role: 'user', content: c.question }, { role: 'assistant', content: a })
+      }
+      return turns
+    },
+    [cases],
+  )
+
+  const run = useCallback(
+    async (id: string, question: string, context: string | null, history: HistoryTurn[]) => {
+      setBusy(true)
+      abortRef.current = new AbortController()
+      try {
+        const res = await fetch('/api/sdk/compliance-advisor/chat', {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({ question, context, history, currentStep, country }),
+          signal: abortRef.current.signal,
+        })
+        if (!res.ok) {
+          const e = await res.json().catch(() => ({ error: 'Unbekannter Fehler' }))
+          throw new Error(e.error || `Server-Fehler (${res.status})`)
+        }
+        const data = (await res.json()) as AdvisorResponse
+        patch(id, { response: data, status: 'done', selectedContext: context })
+      } catch (err) {
+        if ((err as Error).name === 'AbortError') {
+          patch(id, { status: 'done' })
+          return
+        }
+        patch(id, {
+          status: 'error',
+          error: err instanceof Error ? err.message : 'Verbindung fehlgeschlagen',
+        })
+      } finally {
+        setBusy(false)
+      }
+    },
+    [currentStep, country, patch],
+  )
+
+  const ask = useCallback(
+    (question: string, opts?: { newThread?: boolean }) => {
+      const q = question.trim()
+      if (!q || busy) return
+      const startNew = opts?.newThread || !activeThreadId || cases.length === 0
+      const threadId = startNew ? uid('thread') : activeThreadId!
+      const id = uid('case')
+      const history = startNew ? [] : buildHistory(threadId)
+      setCases((prev) => [
+        ...prev,
+        { id, threadId, question: q, response: null, selectedContext: null, status: 'loading' },
+      ])
+      setActiveThreadId(threadId)
+      setActiveCaseId(id)
+      void run(id, q, null, history)
+    },
+    [busy, activeThreadId, cases.length, buildHistory, run],
+  )
+
+  const newTopic = useCallback((question: string) => ask(question, { newThread: true }), [ask])
+
+  const selectContext = useCallback(
+    (id: string, context: string) => {
+      const c = cases.find((x) => x.id === id)
+      if (!c || busy) return
+      patch(id, { status: 'loading', selectedContext: context })
+      void run(id, c.question, context, buildHistory(c.threadId, id))
+    },
+    [cases, busy, run, patch, buildHistory],
+  )
+
+  const remove = useCallback((id: string) => {
+    setCases((prev) => prev.filter((c) => c.id !== id))
+  }, [])
+
+  const selectCase = useCallback((id: string) => {
+    setActiveCaseId(id)
+    setCases((prev) => {
+      const c = prev.find((x) => x.id === id)
+      if (c) setActiveThreadId(c.threadId)
+      return prev
+    })
+  }, [])
+
+  const stop = useCallback(() => {
+    abortRef.current?.abort()
+    setBusy(false)
+  }, [])
+
+  // Keep the active selection valid after deletions.
+  useEffect(() => {
+    if (activeCaseId && !cases.some((c) => c.id === activeCaseId)) {
+      const last = cases[cases.length - 1] ?? null
+      setActiveCaseId(last?.id ?? null)
+      setActiveThreadId(last?.threadId ?? null)
+    }
+  }, [cases, activeCaseId])
+
+  const threads = useMemo<AdvisorThread[]>(() => {
+    const order: string[] = []
+    const byId = new Map<string, AdvisorThread>()
+    for (const c of cases) {
+      let t = byId.get(c.threadId)
+      if (!t) {
+        t = { id: c.threadId, title: c.question, cases: [] }
+        byId.set(c.threadId, t)
+        order.push(c.threadId)
+      }
+      t.cases.push(c)
+    }
+    return order.map((id) => byId.get(id)!)
+  }, [cases])
+
+  return {
+    cases,
+    threads,
+    busy,
+    activeCaseId,
+    activeThreadId,
+    ask,
+    newTopic,
+    selectContext,
+    selectCase,
+    remove,
+    stop,
+  }
+}

admin-compliance/components/sdk/advisor/useAdvisorEmail.ts

@@ -0,0 +1,72 @@
+'use client'
+
+import { useCallback, useState } from 'react'
+import type { AdvisorCase } from './useAdvisorCase'
+
+function esc(s: string): string {
+  return s
+    .replace(/&/g, '&')
+    .replace(/</g, '&lt;')
+    .replace(/>/g, '&gt;')
+    .replace(/"/g, '&quot;')
+}
+
+function evidenceHtml(c: AdvisorCase): string {
+  const ev = c.response?.evidence ?? []
+  if (ev.length === 0) return ''
+  const items = ev
+    .map(
+      (e) =>
+        `<li>${esc(e.document)}${e.section ? ` — ${esc(e.section)}` : ''}${e.paragraph ? ` ${esc(e.paragraph)}` : ''}</li>`,
+    )
+    .join('')
+  return `<p style="color:#64748b;font-size:12px;margin:4px 0 0;">Evidence:</p><ul style="color:#64748b;font-size:12px;margin:2px 0;">${items}</ul>`
+}
+
+/** Sends the consultation cases (question + answer + evidence) as an email to the DSB. */
+export function useAdvisorEmail(cases: AdvisorCase[], country: string, currentStep: string) {
+  const [sending, setSending] = useState(false)
+  const [sent, setSent] = useState(false)
+
+  const send = useCallback(async () => {
+    if (cases.length === 0 || sending) return
+    setSending(true)
+    try {
+      const qaHtml = cases
+        .map((c) => {
+          const a = c.response?.answer || c.response?.general_answer || '(keine Antwort)'
+          return `<div style="margin-bottom:16px;"><p style="font-weight:600;color:#1e293b;">Frage: ${esc(
+            c.question,
+          )}</p><p style="color:#475569;white-space:pre-wrap;">${esc(a)}</p>${evidenceHtml(c)}</div>`
+        })
+        .join('')
+
+      const bodyHtml = `
+        <h2 style="color:#1e293b;">Compliance Advisor — Beratungsprotokoll</h2>
+        <p style="color:#64748b;font-size:13px;">Datum: ${esc(new Date().toLocaleString('de-DE'))} | Land: ${esc(country)} | Kontext: ${esc(currentStep)}</p>
+        <hr style="border-color:#e2e8f0;margin:16px 0;">
+        ${qaHtml}
+        <hr style="border-color:#e2e8f0;margin:16px 0;">
+        <p style="color:#94a3b8;font-size:11px;">Automatisch erstellt vom BreakPilot Compliance Advisor</p>`
+
+      await fetch('/api/sdk/v1/agent/notify', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          recipient: 'dsb@breakpilot.local',
+          subject: `Compliance Advisor — ${cases.length} Fragen (${currentStep})`,
+          body_html: bodyHtml,
+          role: 'Datenschutzbeauftragter',
+        }),
+      })
+      setSent(true)
+      setTimeout(() => setSent(false), 3000)
+    } catch (e) {
+      console.error('Email send failed:', e)
+    } finally {
+      setSending(false)
+    }
+  }, [cases, sending, country, currentStep])
+
+  return { send, sending, sent }
+}

admin-compliance/components/sdk/advisor/useCitationHighlight.ts

@@ -0,0 +1,34 @@
+'use client'
+
+import { useState } from 'react'
+import type { Citation } from '@/lib/sdk/advisor/contract'
+import type { CiteHandler } from './Markdown'
+
+/**
+ * Couples answer [n] markers to evidence cards: clicking [n] highlights + scrolls to the referenced
+ * evidence unit. Works across layout columns via the card's DOM id (ev-<evidence_id>).
+ */
+export function useCitationHighlight(citations: Citation[]): {
+  highlightedId?: string
+  cite?: CiteHandler
+} {
+  const [highlightedId, setHighlightedId] = useState<string | undefined>()
+  if (citations.length === 0) return { highlightedId }
+  return {
+    highlightedId,
+    cite: {
+      count: citations.length,
+      onSelect: (n: number) => {
+        const c = citations[n - 1]
+        if (!c) return
+        setHighlightedId(c.evidence_id)
+        if (typeof document !== 'undefined') {
+          document.getElementById(`ev-${c.evidence_id}`)?.scrollIntoView({
+            behavior: 'smooth',
+            block: 'center',
+          })
+        }
+      },
+    },
+  }
+}

admin-compliance/components/sdk/advisor/useClipboard.ts

@@ -0,0 +1,19 @@
+'use client'
+
+import { useCallback, useState } from 'react'
+
+/** Writes text to the clipboard and flags the copied key briefly (for a check-mark affordance). */
+export function useClipboard(resetMs = 1500) {
+  const [copiedKey, setCopiedKey] = useState<string | null>(null)
+
+  const copy = useCallback(
+    (key: string, text: string) => {
+      void navigator.clipboard?.writeText(text)
+      setCopiedKey(key)
+      window.setTimeout(() => setCopiedKey((k) => (k === key ? null : k)), resetMs)
+    },
+    [resetMs],
+  )
+
+  return { copiedKey, copy }
+}

admin-compliance/e2e/specs/compliance-advisor-widget.spec.ts

@@ -0,0 +1,70 @@
+/**
+ * E2E: Compliance Advisor widget UX — topic threads, new-topic vs follow-up, delete, copy, fullscreen.
+ * Stubs the chat endpoint with an answer fixture so every ask yields a finished case.
+ */
+
+import { test, expect } from '../fixtures/sdk-fixtures'
+
+const CHAT_ROUTE = '**/api/sdk/compliance-advisor/chat'
+const openAdvisor = 'Compliance Advisor oeffnen'
+
+const ANSWER = {
+  mode: 'answer',
+  question: '',
+  clarity: { is_underspecified: false, dominant_context: 'cyber', concentration: 0.9 },
+  general_answer: null,
+  answer: 'Musterantwort [1].',
+  scoped_query: null,
+  evidence: [{ evidence_id: 'e1', document: 'DSGVO', section: 'Art. 5', bindingness: 'binding' }],
+  citations: [{ citation_id: 'c1', number: 1, evidence_id: 'e1', document: 'DSGVO', section: 'Art. 5' }],
+  visual_evidence: [],
+  footnotes: [],
+}
+
+async function stub(page: import('@playwright/test').Page) {
+  await page.route(CHAT_ROUTE, (r) =>
+    r.fulfill({ status: 200, contentType: 'application/json', body: JSON.stringify(ANSWER) }),
+  )
+}
+
+test('new topic creates a second thread; copy control + fullscreen available', async ({ sdkPage }) => {
+  await stub(sdkPage)
+  await sdkPage.getByRole('button', { name: openAdvisor }).click()
+
+  const input = sdkPage.getByPlaceholder('Frage eingeben...')
+  await input.fill('Erste Frage')
+  await input.press('Enter')
+  await expect(sdkPage.getByText(/Musterantwort/)).toBeVisible()
+
+  // expand -> the topic tree ("Themen") appears in the left menu
+  await sdkPage.getByRole('button', { name: 'Vergroessern' }).click()
+  await expect(sdkPage.getByText('Themen')).toBeVisible()
+  await expect(sdkPage.getByText('Erste Frage').first()).toBeVisible()
+
+  // a second, separate topic
+  await sdkPage.getByPlaceholder('Folgefrage eingeben...').fill('Zweites Thema')
+  await sdkPage.getByRole('button', { name: 'Neues Thema' }).click()
+  await expect(sdkPage.getByText('Zweites Thema').first()).toBeVisible()
+  await expect(sdkPage.getByText('Erste Frage').first()).toBeVisible()
+
+  // copy affordance + fullscreen toggle
+  await expect(sdkPage.getByRole('button', { name: 'Diese Frage kopieren' }).first()).toBeVisible()
+  await sdkPage.getByRole('button', { name: 'Vollbild' }).click()
+  await expect(sdkPage.getByRole('button', { name: 'Vollbild verlassen' })).toBeVisible()
+})
+
+test('delete removes a question from the thread', async ({ sdkPage }) => {
+  await stub(sdkPage)
+  await sdkPage.getByRole('button', { name: openAdvisor }).click()
+
+  const input = sdkPage.getByPlaceholder('Frage eingeben...')
+  await input.fill('Zu löschen')
+  await input.press('Enter')
+  await expect(sdkPage.getByText(/Musterantwort/)).toBeVisible()
+
+  await sdkPage.getByRole('button', { name: 'Vergroessern' }).click()
+  await expect(sdkPage.getByText('Zu löschen').first()).toBeVisible()
+
+  await sdkPage.getByRole('button', { name: 'Frage löschen' }).first().click()
+  await expect(sdkPage.getByText('Zu löschen')).toHaveCount(0)
+})

admin-compliance/e2e/specs/compliance-advisor.spec.ts

@@ -0,0 +1,102 @@
+/**
+ * E2E: Compliance Advisor — Clarity Gate (v3 contract)
+ *
+ * Drives the floating advisor widget end-to-end against a stubbed /api/sdk/compliance-advisor/chat
+ * (contract fixtures), so the whole FE chain is exercised without the RAG/LLM backend:
+ *   - underspecified question -> clarify mode (L1 general answer + domain context chips)
+ *   - specific question       -> answer mode (markdown + [n] citation coupling + evidence pane)
+ *   - clarify -> pick a context -> scoped answer
+ * Runs on CI / macmini (needs the Next app on :3002).
+ */
+
+import { test, expect } from '../fixtures/sdk-fixtures'
+
+const CHAT_ROUTE = '**/api/sdk/compliance-advisor/chat'
+const openAdvisor = 'Compliance Advisor oeffnen'
+const inputPlaceholder = 'Frage eingeben...'
+
+const CLARIFY = {
+  mode: 'clarify',
+  question: 'Was ist PDCA?',
+  clarity: {
+    is_underspecified: true,
+    concentration: 0.3,
+    suggested_contexts: [
+      { id: 'datenschutz', label: 'Datenschutz' },
+      { id: 'cyber', label: 'Cybersecurity' },
+    ],
+  },
+  general_answer: 'PDCA steht für **Plan-Do-Check-Act**.',
+  answer: null,
+  evidence: [],
+  citations: [],
+  visual_evidence: [],
+  footnotes: [],
+}
+
+const ANSWER = {
+  mode: 'answer',
+  question: 'CRA Meldefrist',
+  clarity: { is_underspecified: false, dominant_context: 'cyber', concentration: 0.88 },
+  answer: 'Die Meldung erfolgt unverzüglich [1].',
+  evidence: [
+    { evidence_id: 'e1', document: 'CRA', section: 'Art. 14', paragraph: 'Abs. 1', snippet: 'unverzüglich melden', bindingness: 'binding' },
+  ],
+  citations: [
+    { citation_id: 'c1', number: 1, evidence_id: 'e1', document: 'CRA', section: 'Art. 14', paragraph: 'Abs. 1' },
+  ],
+  visual_evidence: [],
+  footnotes: [],
+}
+
+async function ask(page: import('@playwright/test').Page, question: string) {
+  await page.getByRole('button', { name: openAdvisor }).click()
+  const input = page.getByPlaceholder(inputPlaceholder)
+  await input.fill(question)
+  await input.press('Enter')
+}
+
+test.describe('Compliance Advisor — Clarity Gate', () => {
+  test('underspecified question -> clarify (L1 definition + context chips, no evidence)', async ({ sdkPage }) => {
+    await sdkPage.route(CHAT_ROUTE, (r) =>
+      r.fulfill({ status: 200, contentType: 'application/json', body: JSON.stringify(CLARIFY) }),
+    )
+    await ask(sdkPage, 'Was ist PDCA?')
+
+    await expect(sdkPage.getByText('Allgemeine Definition')).toBeVisible()
+    await expect(sdkPage.getByText('Plan-Do-Check-Act')).toBeVisible()
+    await expect(sdkPage.getByRole('button', { name: 'Datenschutz' })).toBeVisible()
+    await expect(sdkPage.getByRole('button', { name: 'Cybersecurity' })).toBeVisible()
+  })
+
+  test('specific question -> answer with [n] citation + evidence pane', async ({ sdkPage }) => {
+    await sdkPage.route(CHAT_ROUTE, (r) =>
+      r.fulfill({ status: 200, contentType: 'application/json', body: JSON.stringify(ANSWER) }),
+    )
+    await ask(sdkPage, 'CRA Meldefrist')
+
+    await expect(sdkPage.getByText(/unverzüglich/)).toBeVisible()
+    await expect(sdkPage.getByTitle('Beleg 1 anzeigen')).toBeVisible()
+    // bindingness present -> header splits into Rechtsgrundlagen vs Leitlinien (evidence framing)
+    await expect(sdkPage.getByText('Rechtsgrundlagen').first()).toBeVisible()
+    // family name resolved for the user (shown both in the summary breakdown and the evidence card)
+    await expect(sdkPage.getByText('Cyber Resilience Act (CRA)').first()).toBeVisible()
+  })
+
+  test('clarify -> pick a context -> scoped answer', async ({ sdkPage }) => {
+    let calls = 0
+    await sdkPage.route(CHAT_ROUTE, (r) => {
+      calls += 1
+      r.fulfill({
+        status: 200,
+        contentType: 'application/json',
+        body: JSON.stringify(calls === 1 ? CLARIFY : ANSWER),
+      })
+    })
+    await ask(sdkPage, 'Was ist PDCA?')
+    await sdkPage.getByRole('button', { name: 'Datenschutz' }).click()
+
+    await expect(sdkPage.getByText(/unverzüglich/)).toBeVisible()
+    await expect(sdkPage.getByTitle('Beleg 1 anzeigen')).toBeVisible()
+  })
+})

admin-compliance/lib/sdk/__tests__/advisor-copy.test.ts

@@ -0,0 +1,65 @@
+import { describe, it, expect } from 'vitest'
+import { formatCaseForCopy, formatThreadForCopy } from '../advisor/copy'
+import type { AdvisorResponse } from '../advisor/contract'
+
+const answer: AdvisorResponse = {
+  mode: 'answer',
+  question: 'CRA Meldefrist',
+  clarity: { is_underspecified: false, concentration: 0.9 },
+  general_answer: null,
+  answer: 'Unverzüglich melden [1].',
+  scoped_query: 'cyber',
+  evidence: [{ evidence_id: 'e1', document: 'CRA', section: 'Art. 14', paragraph: 'Abs. 1', url: 'https://x' }],
+  citations: [],
+  visual_evidence: [],
+  footnotes: [{ footnote_id: 'f1', ref: 'Fußnote 3', document: 'EDPB', section: 'Kap III', text: 'Detail' }],
+}
+
+const clarify: AdvisorResponse = {
+  mode: 'clarify',
+  question: 'Was ist PDCA?',
+  clarity: { is_underspecified: true, concentration: 0.3 },
+  general_answer: 'PDCA = Plan-Do-Check-Act.',
+  answer: null,
+  scoped_query: null,
+  evidence: [],
+  citations: [],
+  visual_evidence: [],
+  footnotes: [],
+}
+
+describe('formatCaseForCopy', () => {
+  it('includes question, answer, resolved evidence and footnotes', () => {
+    const s = formatCaseForCopy({ question: 'CRA Meldefrist', response: answer })
+    expect(s).toContain('### Frage\nCRA Meldefrist')
+    expect(s).toContain('### Antwort\nUnverzüglich melden [1].')
+    expect(s).toContain('### Belege')
+    expect(s).toContain('Cyber Resilience Act (CRA), Art. 14 Abs. 1 (https://x)')
+    expect(s).toContain('### Fußnoten')
+    expect(s).toContain('Fußnote 3 — EDPB / Kap III: Detail')
+  })
+
+  it('falls back to the general definition for a clarify case', () => {
+    const s = formatCaseForCopy({ question: 'Was ist PDCA?', response: clarify })
+    expect(s).toContain('### Antwort\nPDCA = Plan-Do-Check-Act.')
+    expect(s).not.toContain('### Belege')
+  })
+
+  it('handles a case without a response', () => {
+    const s = formatCaseForCopy({ question: 'offen', response: null })
+    expect(s).toContain('### Antwort\n(keine Antwort)')
+  })
+})
+
+describe('formatThreadForCopy', () => {
+  it('renders a title heading + every case separated by a rule', () => {
+    const s = formatThreadForCopy('CRA Meldefrist', [
+      { question: 'CRA Meldefrist', response: answer },
+      { question: 'Und für KMU?', response: clarify },
+    ])
+    expect(s.startsWith('# CRA Meldefrist')).toBe(true)
+    expect(s).toContain('\n---\n')
+    expect(s).toContain('CRA Meldefrist')
+    expect(s).toContain('Und für KMU?')
+  })
+})

admin-compliance/lib/sdk/__tests__/advisor-evidence-grouping.test.ts

@@ -0,0 +1,85 @@
+import { describe, it, expect } from 'vitest'
+import {
+  groupByFamily,
+  provisionSummary,
+  summarizeEvidence,
+  type FamilyGroup,
+} from '../advisor/evidence-grouping'
+import type { EvidenceUnit } from '../advisor/contract'
+
+function u(p: Partial<EvidenceUnit> & { document: string }): EvidenceUnit {
+  return { evidence_id: Math.random().toString(36).slice(2), ...p }
+}
+
+// The Datenschutzerklärung scenario the user reviewed: 6 Kernnormen (5 DSGVO Artikel + § 25 TDDDG)
+// + 2 Leitlinien (DSK, EDPB) across 8 evidence units.
+const DSE: EvidenceUnit[] = [
+  u({ document: 'DSGVO', section: 'Art. 6', bindingness: 'binding' }),
+  u({ document: 'DSGVO', section: 'Art. 7', bindingness: 'binding' }),
+  u({ document: 'DSGVO', section: 'Art. 12', bindingness: 'binding' }),
+  u({ document: 'DSGVO', section: 'Art. 13', bindingness: 'binding' }),
+  u({ document: 'DSGVO', section: 'Art. 14', bindingness: 'binding' }),
+  u({ document: 'TDDDG', section: '§ 25', bindingness: 'binding' }),
+  u({ document: 'DSK', bindingness: 'guidance' }),
+  u({ document: 'EDPB WP 259', bindingness: 'guidance' }),
+]
+
+describe('groupByFamily', () => {
+  it('groups a family and collects distinct provisions in order', () => {
+    const groups = groupByFamily(DSE)
+    const dsgvo = groups.find((g) => g.key === 'dsgvo')!
+    expect(dsgvo.units).toBe(5)
+    expect(dsgvo.sections).toEqual(['Art. 6', 'Art. 7', 'Art. 12', 'Art. 13', 'Art. 14'])
+    expect(dsgvo.bindingness).toBe('binding')
+  })
+
+  it('does not duplicate a repeated section', () => {
+    const groups = groupByFamily([
+      u({ document: 'DSGVO', section: 'Art. 13', bindingness: 'binding' }),
+      u({ document: 'DSGVO', section: 'Art. 13', bindingness: 'binding' }),
+    ])
+    expect(groups[0].sections).toEqual(['Art. 13'])
+    expect(groups[0].units).toBe(2)
+  })
+})
+
+describe('summarizeEvidence', () => {
+  it('splits binding norms from guidance with correct counts', () => {
+    const m = summarizeEvidence(DSE)
+    expect(m.hasBindingness).toBe(true)
+    expect(m.normProvisions).toBe(6) // 5 DSGVO Artikel + § 25 TDDDG
+    expect(m.guidanceCount).toBe(2) // DSK + EDPB
+    expect(m.unitCount).toBe(8)
+    expect(m.norms.map((g) => g.key).sort()).toEqual(['dsgvo', 'tddg'])
+  })
+
+  it('degrades to a neutral breakdown when bindingness is absent', () => {
+    const m = summarizeEvidence([
+      u({ document: 'DSGVO', section: 'Art. 30' }),
+      u({ document: 'CRA', section: 'Art. 14' }),
+    ])
+    expect(m.hasBindingness).toBe(false)
+    expect(m.groups).toHaveLength(2)
+    expect(m.normProvisions).toBe(0)
+    expect(m.guidanceCount).toBe(0)
+  })
+})
+
+describe('provisionSummary', () => {
+  const g = (sections: string[], units = sections.length): FamilyGroup => ({
+    key: 'k',
+    label: 'L',
+    sections,
+    units,
+    bindingness: 'binding',
+  })
+
+  it('names Artikel, §§, single provisions and bare units', () => {
+    expect(provisionSummary(g(['Art. 6', 'Art. 7', 'Art. 13']))).toBe('3 Artikel')
+    expect(provisionSummary(g(['§ 25']))).toBe('§ 25')
+    expect(provisionSummary(g(['§ 25', '§ 26']))).toBe('2 §§')
+    expect(provisionSummary(g(['Art. 13', '§ 25', 'Anhang I']))).toBe('3 Fundstellen')
+    expect(provisionSummary(g([], 3))).toBe('3 Fundstellen')
+    expect(provisionSummary(g([], 1))).toBe('1 Fundstelle')
+  })
+})

admin-compliance/lib/sdk/__tests__/advisor-regulation-display.test.ts

@@ -0,0 +1,31 @@
+import { describe, it, expect } from 'vitest'
+import { resolveRegulation } from '../advisor/regulation-display'
+
+describe('resolveRegulation', () => {
+  it('groups DSK SDM building blocks under one family + extracts the chapter', () => {
+    const b51 = resolveRegulation({ code: 'dsk_sdm_b51', short: 'DSK Sdm B51' })
+    const b41 = resolveRegulation({ code: 'dsk_sdm_b41', short: 'DSK Sdm B41' })
+    const v31 = resolveRegulation({ code: 'dsk_sdm_v31', short: 'DSK Sdm V31' })
+    expect(b51.familyKey).toBe('dsk_sdm')
+    expect(b41.familyKey).toBe('dsk_sdm')
+    expect(v31.familyKey).toBe('dsk_sdm')
+    expect(b51.familyLabel).toContain('Standard-Datenschutzmodell')
+    expect(b51.chapter).toBe('B51')
+    expect(v31.chapter).toBe('V31')
+  })
+
+  it('maps known regulations to friendly family keys', () => {
+    expect(resolveRegulation({ code: 'cra', short: 'CRA' }).familyKey).toBe('cra')
+    expect(resolveRegulation({ code: 'nis2', short: 'NIS2' }).familyKey).toBe('nis2')
+    expect(resolveRegulation({ code: 'dpf', short: 'DPF' }).familyKey).toBe('dpf')
+    expect(resolveRegulation({ code: 'dsgvo', short: 'DS-GVO' }).familyKey).toBe('dsgvo')
+    expect(resolveRegulation({ code: 'bdsg', short: 'BDSG' }).familyKey).toBe('bdsg')
+  })
+
+  it('falls back to code as family + short as label for unknown regulations', () => {
+    const r = resolveRegulation({ code: 'xyz_reg', short: 'XYZ' })
+    expect(r.familyKey).toBe('xyz_reg')
+    expect(r.familyLabel).toBe('XYZ')
+    expect(r.chapter).toBeUndefined()
+  })
+})

admin-compliance/lib/sdk/__tests__/advisor-retrieve-mapping.test.ts

@@ -0,0 +1,81 @@
+import { describe, it, expect } from 'vitest'
+import {
+  resolveMode,
+  mapClarity,
+  mapFootnotes,
+  buildCitations,
+  numberedEvidenceForPrompt,
+  isLegacyRequest,
+} from '../advisor/retrieve-mapping'
+import type { EvidenceUnit } from '../advisor/contract'
+
+describe('resolveMode', () => {
+  it('a chosen context always forces answer', () => expect(resolveMode('clarify', true)).toBe('answer'))
+  it('clarify + no context -> clarify', () => expect(resolveMode('clarify', false)).toBe('clarify'))
+  it('answer -> answer', () => expect(resolveMode('answer', false)).toBe('answer'))
+  it('unknown/undefined -> answer', () => expect(resolveMode(undefined, false)).toBe('answer'))
+})
+
+describe('mapClarity', () => {
+  it('clarify maps candidate_contexts -> suggested_contexts', () => {
+    const c = mapClarity(
+      { mode: 'clarify', concentration: 0.3, candidate_contexts: [{ id: 'ds', label: 'Datenschutz', hits: 5 }] },
+      'clarify',
+    )
+    expect(c.is_underspecified).toBe(true)
+    expect(c.suggested_contexts).toEqual([{ id: 'ds', label: 'Datenschutz' }])
+  })
+  it('answer keeps dominant_context, drops suggestions', () => {
+    const c = mapClarity({ mode: 'answer', concentration: 0.88, dominant_context: 'ds' }, 'answer')
+    expect(c.is_underspecified).toBe(false)
+    expect(c.dominant_context).toBe('ds')
+    expect(c.suggested_contexts).toBeUndefined()
+  })
+})
+
+const ev: EvidenceUnit[] = [
+  { evidence_id: 'e1', document: 'DSGVO', section: 'Art. 30', paragraph: 'Abs. 1', snippet: 'x' },
+  { evidence_id: 'e2', document: 'BDSG', section: '§ 38' },
+]
+
+describe('buildCitations', () => {
+  it('numbers citations 1..n mapped to evidence', () => {
+    const cs = buildCitations(ev)
+    expect(cs).toHaveLength(2)
+    expect(cs[0]).toMatchObject({ citation_id: 'c1', number: 1, evidence_id: 'e1' })
+    expect(cs[1].number).toBe(2)
+  })
+})
+
+describe('numberedEvidenceForPrompt', () => {
+  it('prefixes each unit with [n] + its location', () => {
+    const s = numberedEvidenceForPrompt(ev)
+    expect(s).toContain('[1] DSGVO Art. 30 Abs. 1')
+    expect(s).toContain('[2] BDSG § 38')
+  })
+})
+
+describe('mapFootnotes', () => {
+  it('remaps a /retrieve footnote to the contract footnote', () => {
+    const fns = mapFootnotes([
+      { id: 'f1', number: 17, regulation_short: 'EDPB WP248', section: 'Kap III', text: 't' },
+    ])
+    expect(fns[0]).toMatchObject({
+      footnote_id: 'f1',
+      ref: 'Fußnote 17',
+      document: 'EDPB WP248',
+      section: 'Kap III',
+      text: 't',
+    })
+  })
+})
+
+describe('isLegacyRequest', () => {
+  it('message-only (workspace) -> legacy stream', () => {
+    expect(isLegacyRequest({ message: 'Ist meine DSE ausreichend?' })).toBe(true)
+  })
+  it('question present -> contract (JSON)', () => {
+    expect(isLegacyRequest({ question: 'x', message: 'y' })).toBe(false)
+    expect(isLegacyRequest({ question: 'x' })).toBe(false)
+  })
+})

admin-compliance/lib/sdk/advisor/contract.ts

@@ -0,0 +1,83 @@
+// FE-facing contract for the Compliance Advisor "Case" (Clarity Gate).
+// Matches the SDK<->FE contract (board 2026-07-01 / memory advisor-clarity-gate-contract).
+// The FE renders ONLY these structured fields; it never extracts structure from the answer text.
+// The only exception is rendering the deliberate [n] citation markers, mapped via `citations`.
+
+export interface SuggestedContext {
+  id: string // e.g. "datenschutz"
+  label: string // e.g. "Datenschutz"
+}
+
+export interface ClarityInfo {
+  is_underspecified: boolean
+  concentration: number
+  suggested_contexts?: SuggestedContext[] // clarify mode
+  dominant_context?: string // answer mode
+}
+
+/** A retrieved evidence unit. (`evidence[]` item shape — confirm with SDK; see board rückfrage.) */
+export interface EvidenceUnit {
+  evidence_id: string
+  document: string
+  section?: string
+  paragraph?: string
+  snippet?: string
+  url?: string
+  regulation_code?: string // preferred key for family grouping (from /retrieve)
+  context?: string // knowledge space / domain
+  // Canonical Legal-KG fact (APEX rule): binding norm vs. soft-law guidance. Owned by the
+  // Legal-KG/RAG, not derived in the FE. Absent until /retrieve populates it (board request 2026-07-01);
+  // the FE degrades to a neutral per-regulation breakdown when it is missing.
+  bindingness?: 'binding' | 'guidance'
+}
+
+/** Numbered [n] <-> evidence coupling, produced by the SDK (not parsed from the answer). */
+export interface Citation {
+  citation_id: string
+  number?: number // 1-based marker number ([n])
+  evidence_id: string
+  document: string
+  section?: string | null
+  paragraph?: string | null
+  footnote?: string | null
+  figure?: string | null
+}
+
+/** C8 / visual evidence — `visual_type` generalizes beyond figures (flowchart/bpmn/state_machine/...). */
+export interface VisualEvidence {
+  visual_id: string
+  visual_type: string
+  caption?: string
+  document: string
+  context?: string
+  image_ref?: string
+  vision_summary?: string
+}
+
+export interface Footnote {
+  footnote_id?: string
+  ref?: string
+  document?: string
+  section?: string
+  text?: string
+}
+
+export type AdvisorMode = 'clarify' | 'answer'
+
+export interface AdvisorResponse {
+  mode: AdvisorMode
+  question: string
+  clarity: ClarityInfo
+  general_answer?: string | null // L1 (clarify mode)
+  answer?: string | null // L2 (answer mode)
+  scoped_query?: string | null
+  evidence: EvidenceUnit[]
+  citations: Citation[]
+  visual_evidence: VisualEvidence[]
+  footnotes: Footnote[]
+}
+
+export interface AdvisorRequest {
+  question: string
+  context?: string | null
+}

admin-compliance/lib/sdk/advisor/copy.ts

@@ -0,0 +1,48 @@
+// Pure formatters for "copy to clipboard": a single case (Q + answer + evidence) or a whole thread.
+// No DOM/clipboard here (that is the caller's side effect) so this stays testable.
+
+import type { AdvisorResponse, EvidenceUnit } from './contract'
+import { resolveRegulation } from './regulation-display'
+
+export interface CopyableCase {
+  question: string
+  response: AdvisorResponse | null
+}
+
+function evidenceLine(e: EvidenceUnit): string {
+  const { familyLabel } = resolveRegulation({ code: e.regulation_code || e.document, short: e.document })
+  const loc = [e.section, e.paragraph].filter(Boolean).join(' ')
+  const ref = [familyLabel, loc].filter(Boolean).join(', ')
+  return e.url ? `- ${ref} (${e.url})` : `- ${ref}`
+}
+
+/** One case as portable Markdown: question, answer (or general definition), and its evidence. */
+export function formatCaseForCopy(c: CopyableCase): string {
+  const parts: string[] = [`### Frage`, c.question.trim()]
+  const r = c.response
+  const answer = (r?.answer || r?.general_answer || '').trim()
+  parts.push('', '### Antwort', answer || '(keine Antwort)')
+
+  if (r && r.evidence.length > 0) {
+    parts.push('', '### Belege', ...r.evidence.map(evidenceLine))
+  }
+  if (r && r.footnotes.length > 0) {
+    parts.push(
+      '',
+      '### Fußnoten',
+      ...r.footnotes.map((f, i) => {
+        const head = f.ref || `Fußnote ${i + 1}`
+        const src = [f.document, f.section].filter(Boolean).join(' / ')
+        return `- ${[head, src].filter(Boolean).join(' — ')}${f.text ? `: ${f.text}` : ''}`
+      }),
+    )
+  }
+  return parts.join('\n')
+}
+
+/** A whole topic thread: a title heading followed by every case, separated by rules. */
+export function formatThreadForCopy(title: string, cases: CopyableCase[]): string {
+  const header = `# ${title.trim() || 'Compliance-Advisor-Verlauf'}`
+  const body = cases.map(formatCaseForCopy).join('\n\n---\n\n')
+  return `${header}\n\n${body}`
+}

admin-compliance/lib/sdk/advisor/evidence-grouping.ts

@@ -0,0 +1,80 @@
+// Pure grouping/counting for the "Diese Antwort stützt sich auf" evidence header. No React, testable.
+// Splits evidence into binding norms (Kernnormen) vs. soft-law guidance (Leitlinien) using the
+// Legal-KG-owned `bindingness` fact (APEX rule) — the FE never derives bindingness itself. When the
+// fact is absent it degrades to a neutral per-regulation breakdown (no norm/guidance labels, no
+// fabricated legal classification).
+
+import type { EvidenceUnit } from './contract'
+import { resolveRegulation } from './regulation-display'
+
+export type Bindingness = 'binding' | 'guidance' | 'unknown'
+
+export interface FamilyGroup {
+  key: string // stable family key (grouping)
+  label: string // human-readable regulation name
+  sections: string[] // distinct provisions in first-seen order (e.g. "Art. 13", "§ 25")
+  units: number // raw evidence units in this family
+  bindingness: Bindingness
+}
+
+export interface EvidenceSummaryModel {
+  groups: FamilyGroup[]
+  norms: FamilyGroup[] // bindingness === 'binding'
+  guidance: FamilyGroup[] // bindingness === 'guidance'
+  other: FamilyGroup[] // bindingness unknown
+  hasBindingness: boolean // at least one unit carries the Legal-KG fact
+  normProvisions: number // distinct binding provisions (Kernnormen)
+  guidanceCount: number // distinct guidance documents (Leitlinien)
+  unitCount: number // total evidence units
+}
+
+export function groupByFamily(evidence: EvidenceUnit[]): FamilyGroup[] {
+  const byKey = new Map<string, FamilyGroup>()
+  for (const e of evidence) {
+    const { familyKey, familyLabel } = resolveRegulation({
+      code: e.regulation_code || e.document,
+      short: e.document,
+    })
+    let g = byKey.get(familyKey)
+    if (!g) {
+      g = { key: familyKey, label: familyLabel, sections: [], units: 0, bindingness: 'unknown' }
+      byKey.set(familyKey, g)
+    }
+    g.units += 1
+    if (e.section && !g.sections.includes(e.section)) g.sections.push(e.section)
+    if (e.bindingness && g.bindingness === 'unknown') g.bindingness = e.bindingness
+  }
+  return [...byKey.values()]
+}
+
+/** distinct provisions for a family; falls back to raw unit count when no section is known. */
+export function provisionCount(g: FamilyGroup): number {
+  return g.sections.length || g.units
+}
+
+/** "5 Artikel" / "§ 25" / "3 Fundstellen" — the noun follows the family's own citation style. */
+export function provisionSummary(g: FamilyGroup): string {
+  const n = g.sections.length
+  if (n === 0) return `${g.units} ${g.units === 1 ? 'Fundstelle' : 'Fundstellen'}`
+  if (n === 1) return g.sections[0]
+  if (g.sections.every((s) => /^\s*art/i.test(s))) return `${n} Artikel`
+  if (g.sections.every((s) => s.trim().startsWith('§'))) return `${n} §§`
+  return `${n} Fundstellen`
+}
+
+export function summarizeEvidence(evidence: EvidenceUnit[]): EvidenceSummaryModel {
+  const groups = groupByFamily(evidence)
+  const norms = groups.filter((g) => g.bindingness === 'binding')
+  const guidance = groups.filter((g) => g.bindingness === 'guidance')
+  const other = groups.filter((g) => g.bindingness === 'unknown')
+  return {
+    groups,
+    norms,
+    guidance,
+    other,
+    hasBindingness: norms.length > 0 || guidance.length > 0,
+    normProvisions: norms.reduce((n, g) => n + provisionCount(g), 0),
+    guidanceCount: guidance.length,
+    unitCount: evidence.length,
+  }
+}

admin-compliance/lib/sdk/advisor/regulation-display.ts

@@ -0,0 +1,57 @@
+// Human-readable display for regulations. Maps messy codes/short-names to a stable FAMILY key +
+// friendly label (+ chapter for multi-part works like the DSK SDM). Presentation layer only:
+// it bridges G2 (clean RAG metadata) and keeps working once codes are clean. Extend the table freely.
+
+export interface RegulationRef {
+  code?: string
+  name?: string
+  short?: string
+}
+
+export interface RegulationDisplay {
+  familyKey: string // stable key used to GROUP evidence
+  familyLabel: string // human-readable regulation name
+  chapter?: string // e.g. "B51" for a DSK SDM building block
+}
+
+interface Rule {
+  test: RegExp
+  key: string
+  label: string
+  chapter?: RegExp
+}
+
+// Order matters: more specific patterns first.
+const RULES: Rule[] = [
+  {
+    test: /dsk.?sdm|standard.?datenschutzmodell|(^|[^a-z])sdm([^a-z]|$)/i,
+    key: 'dsk_sdm',
+    label: 'DSK Standard-Datenschutzmodell (SDM)',
+    chapter: /\b([A-Z]\d{1,3})\b/,
+  },
+  { test: /cyber.?resilience|(^|[^a-z])cra([^a-z]|$)/i, key: 'cra', label: 'Cyber Resilience Act (CRA)' },
+  { test: /(^|[^a-z])nis.?2([^a-z]|$)/i, key: 'nis2', label: 'NIS2-Richtlinie' },
+  { test: /data.?privacy.?framework|(^|[^a-z])dpf([^a-z]|$)/i, key: 'dpf', label: 'EU-US Data Privacy Framework' },
+  { test: /maschinen|2023.?1230/i, key: 'maschinenvo', label: 'Maschinenverordnung (EU) 2023/1230' },
+  { test: /ds.?gvo|gdpr/i, key: 'dsgvo', label: 'DSGVO – Datenschutz-Grundverordnung' },
+  { test: /(^|[^a-z])bdsg([^a-z]|$)/i, key: 'bdsg', label: 'BDSG – Bundesdatenschutzgesetz' },
+  { test: /tdddg|ttdsg/i, key: 'tddg', label: 'TDDDG (Digitale-Dienste-Datenschutz)' },
+  { test: /edpb|edsa|(^|[^a-z])wp\s?\d+/i, key: 'edpb', label: 'EDPB / DSK Leitlinien' },
+  { test: /(^|[^a-z])bsi([^a-z]|$)/i, key: 'bsi', label: 'BSI' },
+]
+
+export function resolveRegulation(reg: RegulationRef): RegulationDisplay {
+  const hay = `${reg.code || ''} ${reg.short || ''} ${reg.name || ''}`
+  for (const r of RULES) {
+    if (r.test.test(hay)) {
+      const chapter = r.chapter
+        ? r.chapter.exec(reg.short || reg.code || '')?.[1] || undefined
+        : undefined
+      return { familyKey: r.key, familyLabel: r.label, chapter }
+    }
+  }
+  return {
+    familyKey: reg.code || reg.short || 'unknown',
+    familyLabel: reg.short || reg.name || reg.code || 'Regelwerk',
+  }
+}

admin-compliance/lib/sdk/advisor/retrieve-mapping.ts

@@ -0,0 +1,91 @@
+// Pure mappings from the Go /retrieve response (SDK/RAG-owned; board 2026-07-01 12:25)
+// to the FE-facing advisor contract. Kept pure + testable; the orchestration (route.ts) wires them.
+
+import type { Citation, ClarityInfo, EvidenceUnit, Footnote, VisualEvidence } from './contract'
+
+export interface RetrieveClarity {
+  mode?: string // 'clarify' | 'answer'
+  reason?: string // e.g. 'middle_band_llm_needed'
+  concentration?: number
+  domain_count?: number
+  dominant_context?: string
+  candidate_contexts?: { id: string; label: string; hits?: number }[]
+}
+
+export interface RetrieveFootnote {
+  id?: string
+  ref?: string
+  number?: number
+  regulation_code?: string
+  regulation_short?: string
+  regulation_name?: string
+  section?: string
+  text?: string
+}
+
+export interface RetrieveResponse {
+  evidence?: EvidenceUnit[]
+  visual_evidence?: VisualEvidence[]
+  footnotes?: RetrieveFootnote[]
+  clarity?: RetrieveClarity
+  results?: unknown[]
+  tables?: unknown[] // C6 — not in the FE contract yet (future TablesPane)
+}
+
+/** clarify unless a context was chosen; /retrieve's clarity.mode decides for un-scoped queries. */
+export function resolveMode(clarityMode: string | undefined, hasContext: boolean): 'clarify' | 'answer' {
+  if (hasContext) return 'answer'
+  return clarityMode === 'clarify' ? 'clarify' : 'answer'
+}
+
+export function mapClarity(c: RetrieveClarity | undefined, mode: 'clarify' | 'answer'): ClarityInfo {
+  return {
+    is_underspecified: mode === 'clarify',
+    concentration: c?.concentration ?? 0,
+    dominant_context: c?.dominant_context,
+    suggested_contexts:
+      mode === 'clarify' ? (c?.candidate_contexts ?? []).map((cc) => ({ id: cc.id, label: cc.label })) : undefined,
+  }
+}
+
+export function mapFootnotes(fns: RetrieveFootnote[] | undefined): Footnote[] {
+  return (fns ?? []).map((f) => ({
+    footnote_id: f.id,
+    ref: f.ref ?? (f.number != null ? `Fußnote ${f.number}` : undefined),
+    document: f.regulation_short || f.regulation_name || f.regulation_code,
+    section: f.section,
+    text: f.text,
+  }))
+}
+
+/** Citations are generated by the orchestration (not by /retrieve): [n] -> nth evidence unit. */
+export function buildCitations(evidence: EvidenceUnit[]): Citation[] {
+  return evidence.map((e, i) => ({
+    citation_id: `c${i + 1}`,
+    number: i + 1,
+    evidence_id: e.evidence_id,
+    document: e.document,
+    section: e.section ?? null,
+    paragraph: e.paragraph ?? null,
+    footnote: null,
+    figure: null,
+  }))
+}
+
+/** Numbered evidence list injected into the L2 prompt so the LLM can cite [n]. */
+export function numberedEvidenceForPrompt(evidence: EvidenceUnit[]): string {
+  return evidence
+    .map((e, i) => {
+      const loc = [e.document, e.section, e.paragraph].filter(Boolean).join(' ')
+      return `[${i + 1}] ${loc}\n${e.snippet ?? ''}`.trim()
+    })
+    .join('\n\n')
+}
+
+/**
+ * Backward-compat discriminator: legacy consumers (e.g. breakpilot-workspace) send `{message}`
+ * and read a plain-text stream; the new FE sends `{question}` and expects the JSON contract.
+ */
+export function isLegacyRequest(body: { question?: unknown; message?: unknown }): boolean {
+  return body.question == null && typeof body.message === 'string'
+}

admin-compliance/lib/sdk/agents/advisor-llm.ts

@@ -138,3 +138,26 @@ export async function streamAdvisorAnswer(
   if (ollama) return textStream(ollama, parseOllamaLine)
   return null
 }
+
+/**
+ * Nicht-streamende Variante: sammelt die vollstaendige LLM-Antwort als String (fuer die
+ * JSON-Contract-Antwort der Advisor-Orchestrierung). null = kein LLM erreichbar.
+ */
+export async function completeAdvisorAnswer(messages: ChatMessage[]): Promise<string | null> {
+  const stream = await streamAdvisorAnswer(messages)
+  if (!stream) return null
+  const reader = stream.getReader()
+  const decoder = new TextDecoder()
+  let out = ''
+  try {
+    for (;;) {
+      const { done, value } = await reader.read()
+      if (done) break
+      if (value) out += decoder.decode(value, { stream: true })
+    }
+    out += decoder.decode()
+  } finally {
+    reader.releaseLock()
+  }
+  return out
+}

admin-compliance/lib/sdk/agents/advisor-rag.ts

@@ -7,10 +7,13 @@
  * Advisor bleibt damit collection-agnostisch (Vertrag: Compiler -> Collections -> Retriever
  * -> Advisor); die fruehere Multi-Collection-Logik liegt jetzt im Retriever.
  *
- * Fehler werden geschluckt (graceful: Antwort ohne RAG-Kontext).
- * Fundstellen via article_label sind live ab dem Prod-Re-Ingest 2026-06.
+ * `retrieveAdvisorEvidence` liefert die STRUKTURIERTEN Treffer (fuer das Evidence-Workspace-
+ * Frontend, das nur strukturierte Daten rendert und nie den Antworttext parst) UND den
+ * vorformatierten Kontext-Block fuer den LLM-Prompt. Fehler werden geschluckt (graceful).
  */
 
+import type { RetrieveResponse } from '@/lib/sdk/advisor/retrieve-mapping'
+
 const SDK_URL =
   process.env.SDK_API_URL || process.env.SDK_URL || 'http://ai-compliance-sdk:8090'
 
@@ -18,7 +21,7 @@ const DEFAULT_USER = '00000000-0000-0000-0000-000000000001'
 const DEFAULT_TENANT =
   process.env.DEFAULT_TENANT_ID || '9282a473-5c95-4b3a-bf78-0ecc0ec71d3e'
 
-interface SdkRagResult {
+export interface SdkRagResult {
   text?: string
   regulation_code?: string
   regulation_name?: string
@@ -34,20 +37,27 @@ interface SdkRagResult {
   score?: number
 }
 
+/** Raw RAG response. `figures`/`footnotes` (C8 / C-FN) are passed through untyped until the
+ *  RAG-ingestion contract is finalized (board), then mapped in the evidence-adapter. */
+interface SdkRagResponse {
+  results?: SdkRagResult[]
+  figures?: unknown[]
+  footnotes?: unknown[]
+}
+
 interface ScoredPassage {
   content: string
   source: string
   score: number
 }
 
-/** Normalisiert eine ai-sdk-RAG-Antwort auf {content, source, score}. */
+/** Normalisiert eine ai-sdk-RAG-Antwort auf {content, source, score} (fuer den Prompt-Kontext). */
 export function mapSdkResults(results: SdkRagResult[] | undefined): ScoredPassage[] {
   return (results || [])
     .map((r) => ({
       content: r.text || '',
       // Fundstelle: article_label ist die fertig formatierte, druckbare Quelle aus der
-      // Ingestion ("BDSG § 38 Abs. 1"); Fallback baut sie aus den strukturierten Feldern
-      // (bzw. alt-ingestierte Chunks ohne Legal-Metadaten). Siehe rag_reingest_spec.md §2/§7.
+      // Ingestion ("BDSG § 38 Abs. 1"); Fallback baut sie aus den strukturierten Feldern.
       source:
         (r.article_label && r.article_label.trim()) ||
         [r.regulation_short || r.regulation_name || r.regulation_code, r.article, r.paragraph, r.sub]
@@ -59,15 +69,16 @@ export function mapSdkResults(results: SdkRagResult[] | undefined): ScoredPassag
     .filter((p) => p.content)
 }
 
-/**
- * Authority Router: EIN collection-agnostischer Aufruf an die ai-sdk (`/sdk/v1/rag/retrieve`).
- * Der Router waehlt die Collections (Broad-Authority-Base + KB-2026.1-Slice bei in-scope),
- * merged + authority-ranked sie und liefert die Top-Passagen. Der Advisor weiss damit nichts
- * mehr ueber einzelne Collections — die fruehere Multi-Collection-Logik liegt jetzt im Retriever.
- * Fehler werden geschluckt (graceful: Antwort ohne RAG-Kontext).
- */
-export async function queryAdvisorRAG(query: string): Promise<string> {
-  let passages: ScoredPassage[] = []
+/** Formatiert die Top-Passagen als Kontext-Block fuer den System-Prompt. */
+function formatContext(passages: ScoredPassage[]): string {
+  if (passages.length === 0) return ''
+  return passages
+    .map((r, i) => `[Quelle ${i + 1}: ${r.source}]\n${r.content}`)
+    .join('\n\n---\n\n')
+}
+
+/** EIN collection-agnostischer Aufruf an die ai-sdk. Fehler -> leeres Ergebnis (graceful). */
+async function fetchRag(query: string): Promise<SdkRagResponse> {
   try {
     const res = await fetch(`${SDK_URL}/sdk/v1/rag/retrieve`, {
       method: 'POST',
@@ -79,16 +90,61 @@ export async function queryAdvisorRAG(query: string): Promise<string> {
       body: JSON.stringify({ query, top_k: 8 }),
       signal: AbortSignal.timeout(15000),
     })
-    if (res.ok) {
-      const data = await res.json()
-      passages = mapSdkResults(data.results)
-    }
+    if (res.ok) return ((await res.json()) as SdkRagResponse) || {}
   } catch {
     // graceful: keine Verbindung -> Antwort ohne RAG-Kontext
   }
-  // Der Router liefert bereits authority-geordnete Top-K; Reihenfolge bewahren.
-  if (passages.length === 0) return ''
-  return passages
-    .map((r, i) => `[Quelle ${i + 1}: ${r.source}]\n${r.content}`)
-    .join('\n\n---\n\n')
+  return {}
+}
+
+export interface AdvisorEvidenceRaw {
+  contextText: string
+  results: SdkRagResult[]
+  figures?: unknown[]
+  footnotes?: unknown[]
+}
+
+/**
+ * Strukturierte Evidence + Prompt-Kontext aus EINEM Retrieval. Das Frontend bekommt die
+ * `results` (und kuenftig `figures`/`footnotes`) als Daten; der `contextText` geht in den
+ * LLM-Prompt. Reihenfolge der authority-geordneten Top-K bleibt erhalten.
+ */
+export async function retrieveAdvisorEvidence(query: string): Promise<AdvisorEvidenceRaw> {
+  const data = await fetchRag(query)
+  const results = data.results || []
+  return {
+    contextText: formatContext(mapSdkResults(results)),
+    results,
+    figures: Array.isArray(data.figures) ? data.figures : undefined,
+    footnotes: Array.isArray(data.footnotes) ? data.footnotes : undefined,
+  }
+}
+
+/** Abwaertskompatibel: nur der Prompt-Kontext als String. */
+export async function queryAdvisorRAG(query: string): Promise<string> {
+  return (await retrieveAdvisorEvidence(query)).contextText
+}
+
+/**
+ * Voller `/retrieve`-Aufruf fuer die Clarity-Gate-Orchestrierung: liefert die strukturierte
+ * SDK/RAG-Response (evidence/visual_evidence/footnotes/tables/clarity/results). `context` scopet
+ * den 2. Aufruf auf die gewaehlte Domaene. Fehler -> leeres Ergebnis (graceful).
+ */
+export async function retrieveFull(query: string, context?: string | null): Promise<RetrieveResponse> {
+  try {
+    const res = await fetch(`${SDK_URL}/sdk/v1/rag/retrieve`, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'X-User-ID': DEFAULT_USER,
+        'X-Tenant-ID': DEFAULT_TENANT,
+      },
+      body: JSON.stringify({ query, top_k: 8, ...(context ? { context } : {}) }),
+      signal: AbortSignal.timeout(15000),
+    })
+    if (res.ok) return ((await res.json()) as RetrieveResponse) || {}
+  } catch {
+    // graceful: keine Verbindung -> leeres Ergebnis
+  }
+  return {}
 }

ai-compliance-sdk/internal/api/handlers/rag_handlers.go

@@ -1,6 +1,9 @@
 package handlers
 
 import (
+	"encoding/json"
+	"fmt"
+	"log"
 	"net/http"
 	"strconv"
 
@@ -87,6 +90,7 @@ func (h *RAGHandlers) Search(c *gin.Context) {
 type RetrieveRequest struct {
 	Query string `json:"query" binding:"required"`
 	TopK  int    `json:"top_k,omitempty"`
+	Context string `json:"context,omitempty"`
 }
 
 // Retrieve is the Authority Router endpoint. The Advisor calls this with ONLY a query and stays
@@ -105,20 +109,171 @@ func (h *RAGHandlers) Retrieve(c *gin.Context) {
 		req.TopK = 8
 	}
 
+	// E2 Term Resolution: expand unambiguous abbreviations (TOM/VVT/AVV/DSB/DSFA) into the
+	// query so retrieval finds them; ambiguous ones (DSE/DPA) are surfaced to the FE — NOT
+	// auto-mapped (chat context E1 wins, else the FE asks).
+	intent := ucca.DetectIntent(req.Query)
+	termRes := ucca.ResolveAbbreviations(req.Query)
+	req.Query = termRes.Expanded
+
 	results, err := h.ragClient.Retrieve(c.Request.Context(), req.Query, req.TopK)
 	if err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": "RAG retrieve failed: " + err.Error()})
 		return
 	}
 
+	// Evidence-Type-Schicht: die autoritative typisierte Evidence (Fußnoten/Tabellen/Figuren) aus
+	// dem KB-Wissensraum SEPARAT surfacen, statt sie im Breit-Basis-Text-Merge zu verlieren.
+	// results[] bleibt der Text-Kontext fürs LLM + die Quellen-Liste.
+	// Context scoping (E5): the user explicitly chose a knowledge space (chip), so scope
+	// the evidence HARD to it (wider re-retrieve + domain filter) — no off-domain regelwerke
+	// (MDR/UStG/eIDAS) after a context decision.
+	if req.Context != "" {
+		if wide, werr := h.ragClient.Retrieve(c.Request.Context(), req.Query, 30); werr == nil && len(wide) > 0 {
+			results = ucca.FilterByKnowledgeSpace(wide, req.Context, req.TopK)
+		} else {
+			results = ucca.FilterByKnowledgeSpace(results, req.Context, req.TopK)
+		}
+	}
+
+	// G1 scope-gating: a named regulation scopes the evidence to its knowledge space.
+	// Re-retrieve wider and lead with the named regulation's domain so the L2 answer +
+	// [n] citations are built on scoped evidence, not the embedding-majority domain.
+	if scope := ucca.QueryKnowledgeSpace(req.Query); scope != "" {
+		if wide, werr := h.ragClient.Retrieve(c.Request.Context(), req.Query, 30); werr == nil && len(wide) > 0 {
+			results = ucca.ScopeResults(wide, scope, req.TopK)
+		} else {
+			results = ucca.ScopeResults(results, scope, req.TopK)
+		}
+	}
+
+	ev := h.ragClient.RetrieveEvidence(c.Request.Context(), req.Query)
+	// Concept->Norm recall injector: if the query names a legal concept, fetch its
+	// load-bearing norms (Datenschutzerklärung -> Art. 12/13/14 DSGVO, ...) and inject
+	// them into the evidence set so they surface (embedding similarity misses them).
+	if norms := ucca.ConceptNorms(req.Query); len(norms) > 0 {
+		top := 0.9
+		if len(results) > 0 {
+			top = results[0].Score
+		}
+		injected := h.ragClient.FetchByNormIDs(c.Request.Context(), norms, top-0.001)
+		results = ucca.InjectConceptNorms(results, injected, req.TopK)
+	}
+	clarity := ucca.ClassifyClarity(req.Query, results)
+	traceClarity(req.Query, clarity, results)
+
 	c.JSON(http.StatusOK, gin.H{
 		"query":      req.Query,
 		"results":    results,
 		"count":      len(results),
 		"assessment": ucca.Assess(results),
+		"footnotes":  footnotesFromEvidence(ev[ucca.EvidenceFootnote]),
+		"tables":     tablesFromEvidence(ev[ucca.EvidenceTable]),
+		"evidence":        evidenceFromResults(results),
+		"visual_evidence": visualEvidenceFromEvidence(ev[ucca.EvidenceFigure]),
+		"clarity":   clarity,
+		"term_resolution": termRes.Ambiguous,
+		"interaction_intent": intent,
 	})
 }
 
+// footnotesFromEvidence maps FOOTNOTE evidence to the Evidence-Workspace RawFootnote shape.
+func footnotesFromEvidence(rs []ucca.LegalSearchResult) []gin.H {
+	out := make([]gin.H, 0, len(rs))
+	for _, r := range rs {
+		out = append(out, gin.H{
+			"id":               r.CitationUnit,
+			"ref":              r.CitationUnit,
+			"number":           r.FootnoteLabel,
+			"regulation_code":  r.RegulationCode,
+			"regulation_short": r.RegulationShort,
+			"regulation_name":  r.RegulationName,
+			"section":          r.RefCitationUnit,
+			"text":             r.FootnoteVerbatim,
+		})
+	}
+	return out
+}
+
+// tablesFromEvidence maps TABLE evidence (C6/C9). Key is present so the same Evidence-Type path
+// carries tables the moment the UI adds a table section.
+func tablesFromEvidence(rs []ucca.LegalSearchResult) []gin.H {
+	out := make([]gin.H, 0, len(rs))
+	for _, r := range rs {
+		out = append(out, gin.H{
+			"id":               r.CitationUnit,
+			"caption":          r.ArticleLabel,
+			"regulation_code":  r.RegulationCode,
+			"regulation_short": r.RegulationShort,
+			"regulation_name":  r.RegulationName,
+			"section":          r.RefCitationUnit,
+			"text":             r.Text,
+		})
+	}
+	return out
+}
+
+// visualEvidenceFromEvidence maps FIGURE evidence to the Visual Evidence contract shape
+// (C8). visual_type/image_ref/vision_summary populate once C8 lands; the shape is stable now.
+func visualEvidenceFromEvidence(rs []ucca.LegalSearchResult) []gin.H {
+	out := make([]gin.H, 0, len(rs))
+	for _, r := range rs {
+		out = append(out, gin.H{
+			"visual_id":       r.CitationUnit,
+			"visual_type":     "figure",
+			"caption":         r.ArticleLabel,
+			"document":        evidenceDocName(r),
+			"context":         ucca.KnowledgeSpaceOf(r.RegulationCode),
+			"regulation_code": r.RegulationCode,
+			"section":         r.RefCitationUnit,
+			"image_ref":       "",
+			"vision_summary":  "",
+		})
+	}
+	return out
+}
+
+// evidenceFromResults maps retrieval hits to the Evidence contract shape the Advisor
+// Evidence Workspace renders (citations[] reference evidence_id). Populated at retrieve
+// time; citations[] (the [n]<->evidence coupling) come from the answer-generation step.
+func evidenceFromResults(rs []ucca.LegalSearchResult) []gin.H {
+	out := make([]gin.H, 0, len(rs))
+	for _, r := range rs {
+		id := r.CitationUnit
+		if id == "" {
+			id = r.ArticleLabel
+		}
+		out = append(out, gin.H{
+			"evidence_id":     id,
+			"document":        evidenceDocName(r),
+			"section":         r.ArticleLabel,
+			"paragraph":       r.Paragraph,
+			"snippet":         evidenceSnippet(r.Text, 280),
+			"url":             r.SourceURL,
+			"regulation_code": r.RegulationCode,
+			"context":         ucca.KnowledgeSpaceOf(r.RegulationCode),
+		})
+	}
+	return out
+}
+
+// evidenceDocName is the human-facing source name (short code, else full name).
+func evidenceDocName(r ucca.LegalSearchResult) string {
+	if r.RegulationShort != "" {
+		return r.RegulationShort
+	}
+	return r.RegulationName
+}
+
+// evidenceSnippet returns a trimmed excerpt of at most n runes.
+func evidenceSnippet(s string, n int) string {
+	rs := []rune(s)
+	if len(rs) <= n {
+		return s
+	}
+	return string(rs[:n]) + "…"
+}
+
 // ListRegulations returns the list of available regulations in the corpus.
 // GET /sdk/v1/rag/regulations
 func (h *RAGHandlers) ListRegulations(c *gin.Context) {
@@ -273,3 +428,29 @@ func (h *RAGHandlers) LegalCorpusStructure(c *gin.Context) {
 		},
 	})
 }
+
+// traceClarity emits a structured CLARITY_TRACE log line per retrieve for the macmini
+// test session, so qualitative user ratings can be correlated with the gate decision.
+func traceClarity(query string, cl ucca.Clarity, results []ucca.LegalSearchResult) {
+	top := make([]string, 0, 3)
+	for i, r := range results {
+		if i >= 3 {
+			break
+		}
+		top = append(top, r.RegulationShort)
+	}
+	chips := make([]string, 0, len(cl.CandidateContexts))
+	for _, c := range cl.CandidateContexts {
+		chips = append(chips, fmt.Sprintf("%s:%d", c.ID, c.Hits))
+	}
+	b, _ := json.Marshal(map[string]interface{}{
+		"query":         query,
+		"mode":          cl.Mode,
+		"reason":        cl.Reason,
+		"concentration": cl.Concentration,
+		"dominant":      cl.DominantContext,
+		"chips":         chips,
+		"top_evidence":  top,
+	})
+	log.Printf("CLARITY_TRACE %s", string(b))
+}

ai-compliance-sdk/internal/ucca/abbrev_test.go

@@ -0,0 +1,33 @@
+package ucca
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestResolveAbbreviations(t *testing.T) {
+	// unambiguous -> expanded, not flagged
+	tr := ResolveAbbreviations("Was ist eine TOM?")
+	if !strings.Contains(tr.Expanded, "technische und organisatorische") {
+		t.Errorf("TOM must be expanded, got %q", tr.Expanded)
+	}
+	if len(tr.Ambiguous) != 0 {
+		t.Errorf("TOM must not be ambiguous, got %v", tr.Ambiguous)
+	}
+	// ambiguous DSE -> flagged, NOT auto-expanded (chat context must win, else FE asks)
+	tr2 := ResolveAbbreviations("welche Infos in eine DSE?")
+	if tr2.Expanded != "welche Infos in eine DSE?" {
+		t.Errorf("DSE must NOT be auto-mapped, got %q", tr2.Expanded)
+	}
+	if len(tr2.Ambiguous) != 1 || tr2.Ambiguous[0].Abbreviation != "DSE" || len(tr2.Ambiguous[0].Candidates) != 2 {
+		t.Errorf("DSE must be flagged ambiguous with 2 candidates, got %v", tr2.Ambiguous)
+	}
+	// no abbreviation -> unchanged
+	if ResolveAbbreviations("Wie ist das Wetter?").Expanded != "Wie ist das Wetter?" {
+		t.Errorf("query without abbreviation must be unchanged")
+	}
+	// substring must NOT match ("atom" contains "tom" but is not the word TOM)
+	if strings.Contains(ResolveAbbreviations("Was ist ein Atom?").Expanded, "organisatorische") {
+		t.Errorf("substring 'tom' in 'Atom' must not trigger expansion")
+	}
+}

ai-compliance-sdk/internal/ucca/abbreviation_lexicon.go

@@ -0,0 +1,65 @@
+package ucca
+
+import (
+	"strings"
+	"unicode"
+)
+
+// TermResolution is the E2 (Term Resolution) signal in the Advisor Reasoning Stack.
+// Expanded drives retrieval internally (unambiguous abbreviations are spelled out so
+// the embedding/concept layer finds them). Ambiguous is surfaced to the FE, which
+// resolves it via chat context (E1) or asks the user ("Meinst du X oder Y?"). The
+// lexicon NEVER auto-maps an ambiguous abbreviation (e.g. DSE) — real-life discipline.
+type TermResolution struct {
+	Expanded  string          `json:"-"`
+	Ambiguous []TermAmbiguity `json:"ambiguous,omitempty"`
+}
+
+// TermAmbiguity flags one abbreviation the SDK could not resolve deterministically.
+type TermAmbiguity struct {
+	Abbreviation string   `json:"abbreviation"`
+	Candidates   []string `json:"candidates"`
+}
+
+// abbreviationLexicon maps a (lowercased) abbreviation to its canonical term(s).
+// >1 candidate = ambiguous → flagged, not expanded. Start small (User-Spec).
+var abbreviationLexicon = map[string][]string{
+	"dse":  {"Datenschutzerklärung", "Datenschutz-Folgenabschätzung"}, // ambiguous — context wins, else ask
+	"dsfa": {"Datenschutz-Folgenabschätzung"},
+	"tom":  {"technische und organisatorische Maßnahmen"},
+	"vvt":  {"Verzeichnis von Verarbeitungstätigkeiten"},
+	"avv":  {"Auftragsverarbeitungsvertrag"},
+	"dsb":  {"Datenschutzbeauftragter"},
+	"dpa":  {"Data Processing Agreement", "Datenschutzaufsichtsbehörde"}, // ambiguous
+}
+
+// ResolveAbbreviations expands unambiguous abbreviations into the query and flags
+// ambiguous ones. Deterministic: iterates query tokens in order (no map-order
+// dependence). Whole-word match (case-insensitive) so "TOM" hits but "atom" does not.
+func ResolveAbbreviations(query string) TermResolution {
+	tr := TermResolution{Expanded: query}
+	words := strings.FieldsFunc(query, func(r rune) bool {
+		return !unicode.IsLetter(r) && !unicode.IsNumber(r)
+	})
+	seen := map[string]bool{}
+	var expansions []string
+	for _, w := range words {
+		lw := strings.ToLower(w)
+		cands, ok := abbreviationLexicon[lw]
+		if !ok || seen[lw] {
+			continue
+		}
+		seen[lw] = true
+		if len(cands) == 1 {
+			expansions = append(expansions, cands[0])
+		} else {
+			tr.Ambiguous = append(tr.Ambiguous, TermAmbiguity{
+				Abbreviation: strings.ToUpper(lw), Candidates: cands,
+			})
+		}
+	}
+	if len(expansions) > 0 {
+		tr.Expanded = query + " " + strings.Join(expansions, " ")
+	}
+	return tr
+}

ai-compliance-sdk/internal/ucca/authority_router.go

@@ -74,6 +74,10 @@ func (c *LegalRAGClient) Retrieve(ctx context.Context, query string, topK int) (
 		}
 	}
 
+	// Embed the query ONCE and stash it in ctx so the concurrent per-collection searches
+	// below reuse it instead of each re-embedding (was N remote round-trips on dev/OVH).
+	ctx = c.withQueryEmbedding(ctx, query)
+
 	out := make([][]LegalSearchResult, len(collections))
 	var wg sync.WaitGroup
 	for i, coll := range collections {

ai-compliance-sdk/internal/ucca/clarity.go

@@ -0,0 +1,135 @@
+package ucca
+
+import (
+	"sort"
+	"strings"
+)
+
+// Clarity is the READ-ONLY, INSTRUMENTED clarity-gate signal emitted alongside a
+// retrieve response. It does NOT change retrieval or advisor behaviour yet — the
+// advisor still answers normally. Once ~30-50 real questions are collected the
+// thresholds get finalised and the gate is activated in the advisor flow.
+//
+// Ambiguity has two independent sources (empirically measured, 12-question set):
+//   - retrieval scatter: hits spread across many knowledge spaces (low
+//     concentration / high domain_count) — the retriever itself can't localise.
+//   - conceptual generality: a general term the corpus OVER-localises (e.g. "PDCA"
+//     concentrates on datenschutz but is cross-domain) — only an LLM knows this.
+//     The middle band is where the LLM-intent classifier must decide.
+//
+// G1 (explicit scope): when the query NAMES a regulation ("... nach TRGS", "CRA
+// ...", "MaschinenVO ..."), that explicit context beats the embedding scatter —
+// the gate scopes to the named regulation's knowledge space regardless of
+// concentration. This is regulation detection, NOT a broad-term list.
+type Clarity struct {
+	Mode              string           `json:"mode"`               // "answer" | "clarify"
+	Reason            string           `json:"reason"`             // low_concentration | many_domains | high_confidence_scope | middle_band_llm_needed | explicit_scope | no_domain_signal
+	Concentration     float64          `json:"concentration"`      // fraction of tagged hits in the dominant knowledge space
+	DomainCount       int              `json:"domain_count"`       // distinct knowledge spaces in the hits
+	DominantContext   string           `json:"dominant_context"`   // knowledge-space id (explicit scope wins if the query names a regulation)
+	CandidateContexts []ClarityContext `json:"candidate_contexts"` // corpus-grounded chips (spaces actually present)
+}
+
+// ClarityContext is one corpus-grounded context chip.
+type ClarityContext struct {
+	ID    string `json:"id"`
+	Label string `json:"label"`
+	Hits  int    `json:"hits"`
+}
+
+// Tiered thresholds — INSTRUMENTED DEFAULTS, calibrate on 30-50 real questions.
+const (
+	clarityMaxConcentration = 0.45 // <= this => clarify (retrieval scatter)
+	clarityMinDomains       = 4    // >= this => clarify (broad spread)
+	clarityAnswerConc       = 0.75 // >= this => answer (confident scope)
+)
+
+// QueryKnowledgeSpace detects an EXPLICIT regulation mention in the query and maps
+// it to a knowledge space. Regulation detection (authority), not a broad-term list:
+// only fires when the user names a concrete regelwerk. "" if none named.
+func QueryKnowledgeSpace(query string) string {
+	q := " " + strings.ToLower(query) + " "
+	has := func(subs ...string) bool {
+		for _, s := range subs {
+			if strings.Contains(q, s) {
+				return true
+			}
+		}
+		return false
+	}
+	switch {
+	case has("trgs", "trbs", " asr ", "gefahrstoff", "arbeitsplatzgrenzwert", "arbeitsschutz"):
+		return "arbeitsschutz"
+	case has("dsgvo", "gdpr", "bdsg", "tdddg", "ttdsg", " dsk ", "edpb", "datenschutz", " dsfa "):
+		return "datenschutz"
+	case has(" cra ", "cyber resilience", "nis2", "nis-2", " dora ", "enisa", "bsig", "kritis"):
+		return "cyber"
+	case has("ai act", "ki-vo", "ki-verordnung", "ki-system"):
+		return "ki"
+	case has("maschinenverordnung", "maschinenvo", "maschvo", "maschinenrichtlinie", " gpsr ", "produktsicherheit"):
+		return "produktsicherheit"
+	case has(" mdr ", "medizinprodukt", "medical device"):
+		return "produktsicherheit"
+	default:
+		return ""
+	}
+}
+
+// ClassifyClarity computes the read-only clarity signal. Deterministic tiers on the
+// knowledge-space concentration, PLUS the G1 explicit-scope override: if the query
+// names a regulation, that scope wins over the embedding scatter.
+func ClassifyClarity(query string, results []LegalSearchResult) Clarity {
+	counts := map[string]int{}
+	total := 0
+	for _, r := range results {
+		if s := KnowledgeSpaceOf(r.RegulationCode); s != "" {
+			counts[s]++
+			total++
+		}
+	}
+	cl := Clarity{Mode: "answer", Reason: "high_confidence_scope", CandidateContexts: []ClarityContext{}}
+	if total == 0 {
+		cl.Mode, cl.Reason = "clarify", "no_domain_signal"
+		if ks := QueryKnowledgeSpace(query); ks != "" {
+			cl.Mode, cl.Reason, cl.DominantContext = "answer", "explicit_scope", ks
+		}
+		return cl
+	}
+	type kc struct {
+		id string
+		n  int
+	}
+	ks := make([]kc, 0, len(counts))
+	for id, n := range counts {
+		ks = append(ks, kc{id, n})
+	}
+	sort.Slice(ks, func(i, j int) bool {
+		if ks[i].n != ks[j].n {
+			return ks[i].n > ks[j].n
+		}
+		return ks[i].id < ks[j].id
+	})
+	cl.DominantContext = ks[0].id
+	cl.Concentration = float64(ks[0].n) / float64(total)
+	cl.DomainCount = len(counts)
+	for _, k := range ks {
+		cl.CandidateContexts = append(cl.CandidateContexts, ClarityContext{
+			ID: k.id, Label: KnowledgeSpaceLabel[k.id], Hits: k.n,
+		})
+	}
+	switch {
+	case cl.Concentration <= clarityMaxConcentration:
+		cl.Mode, cl.Reason = "clarify", "low_concentration"
+	case cl.DomainCount >= clarityMinDomains:
+		cl.Mode, cl.Reason = "clarify", "many_domains"
+	case cl.Concentration >= clarityAnswerConc:
+		cl.Mode, cl.Reason = "answer", "high_confidence_scope"
+	default:
+		cl.Mode, cl.Reason = "answer", "middle_band_llm_needed"
+	}
+	// G1: an explicitly named regulation beats the embedding scatter.
+	if q := QueryKnowledgeSpace(query); q != "" {
+		cl.Mode, cl.Reason, cl.DominantContext = "answer", "explicit_scope", q
+	}
+	return cl
+}

ai-compliance-sdk/internal/ucca/clarity_test.go

@@ -0,0 +1,64 @@
+package ucca
+
+import "testing"
+
+func TestKnowledgeSpaceOf(t *testing.T) {
+	cases := map[string]string{
+		"DSGVO":                "datenschutz",
+		"BDSG":                 "datenschutz",
+		"DSK SDM B51 ZUGRIFFE": "datenschutz",
+		"EDPS DIGITAL ETHICS":  "datenschutz",
+		"TRGS 900":             "arbeitsschutz",
+		"OSHA 1910 SUBPART O":  "arbeitsschutz",
+		"HGB":                  "wirtschaftsrecht",
+		"BGB":                  "wirtschaftsrecht",
+		"MASCHINENVO":          "produktsicherheit",
+		"MVO":                  "produktsicherheit",
+		"CRA":                  "cyber",
+		"NIST SP800 53R5":      "cyber",
+		"AI ACT":               "ki",
+		"KI-VO":                "ki",
+		"DORA":                 "finanz",
+		"ARG":                  "arbeitsrecht",
+		"":                     "",
+	}
+	for code, want := range cases {
+		if got := KnowledgeSpaceOf(code); got != want {
+			t.Errorf("KnowledgeSpaceOf(%q)=%q want %q", code, got, want)
+		}
+	}
+}
+
+func TestClassifyClarity(t *testing.T) {
+	scattered := []LegalSearchResult{
+		{RegulationCode: "CRA"}, {RegulationCode: "MASCHINENVO"}, {RegulationCode: "EU MDR"},
+		{RegulationCode: "KI-VO"}, {RegulationCode: "TRBS 1111"}, {RegulationCode: "OWASP TOP10"},
+	}
+	if c := ClassifyClarity("Welche Risiken gibt es?", scattered); c.Mode != "clarify" {
+		t.Errorf("scattered: mode=%q reason=%q want clarify", c.Mode, c.Reason)
+	}
+	concentrated := []LegalSearchResult{
+		{RegulationCode: "DSGVO"}, {RegulationCode: "BDSG"}, {RegulationCode: "DSK SDM"},
+		{RegulationCode: "EDPB WP243"}, {RegulationCode: "TDDDG"},
+	}
+	c := ClassifyClarity("Was ist eine DSFA?", concentrated)
+	if c.Mode != "answer" || c.DominantContext != "datenschutz" {
+		t.Errorf("concentrated: mode=%q dominant=%q want answer/datenschutz", c.Mode, c.DominantContext)
+	}
+}
+
+func TestClassifyClarity_ExplicitScope(t *testing.T) {
+	// G1: query names TRGS -> arbeitsschutz wins even though retrieval scatters to datenschutz.
+	scattered := []LegalSearchResult{
+		{RegulationCode: "DSK SDM METHODE"}, {RegulationCode: "DSK SDM V31"}, {RegulationCode: "DSK SDM B41 PLANEN"},
+		{RegulationCode: "DSGVO"}, {RegulationCode: "DSK SDM"}, {RegulationCode: "TRGS 900"}, {RegulationCode: "TRGS 554"},
+	}
+	c := ClassifyClarity("Schwellwertanalyse nach TRGS", scattered)
+	if c.Mode != "answer" || c.Reason != "explicit_scope" || c.DominantContext != "arbeitsschutz" {
+		t.Errorf("explicit TRGS: mode=%q reason=%q dominant=%q want answer/explicit_scope/arbeitsschutz", c.Mode, c.Reason, c.DominantContext)
+	}
+	// no regulation named -> falls through to tiered logic
+	if c := ClassifyClarity("Welche Risiken gibt es?", scattered); c.Reason == "explicit_scope" {
+		t.Errorf("no reg named should not be explicit_scope, got %q", c.Reason)
+	}
+}

ai-compliance-sdk/internal/ucca/concept_fetch.go

@@ -0,0 +1,88 @@
+package ucca
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+)
+
+// FetchByNormIDs loads one representative unit per norm_id from the KB slice
+// collection — the fetch side of the Concept->Norm recall injector. Returns
+// LegalSearchResult with the caller-provided concept-relevance score (there is no
+// similarity query; the injector places them by that score). Returns nil on any
+// error or when no KB slice is configured (graceful degradation).
+func (c *LegalRAGClient) FetchByNormIDs(ctx context.Context, normIDs []string, score float64) []LegalSearchResult {
+	if c.kbSliceCollection == "" || len(normIDs) == 0 {
+		return nil
+	}
+	should := make([]map[string]interface{}, 0, len(normIDs))
+	for _, nid := range normIDs {
+		should = append(should, map[string]interface{}{"key": "norm_id", "match": map[string]interface{}{"value": nid}})
+	}
+	reqBody := map[string]interface{}{
+		"limit":        len(normIDs) * 3,
+		"with_payload": true,
+		"with_vectors": false,
+		"filter":       map[string]interface{}{"should": should},
+	}
+	jsonBody, err := json.Marshal(reqBody)
+	if err != nil {
+		return nil
+	}
+	url := fmt.Sprintf("%s/collections/%s/points/scroll", c.qdrantURL, c.kbSliceCollection)
+	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(jsonBody))
+	if err != nil {
+		return nil
+	}
+	req.Header.Set("Content-Type", "application/json")
+	if c.qdrantAPIKey != "" {
+		req.Header.Set("api-key", c.qdrantAPIKey)
+	}
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil
+	}
+	defer func() { _ = resp.Body.Close() }()
+	if resp.StatusCode != http.StatusOK {
+		return nil
+	}
+	var scrollResp qdrantScrollResponse
+	if err := json.NewDecoder(resp.Body).Decode(&scrollResp); err != nil {
+		return nil
+	}
+	seen := map[string]bool{}
+	out := make([]LegalSearchResult, 0, len(normIDs))
+	for _, pt := range scrollResp.Result.Points {
+		nid := getString(pt.Payload, "norm_id")
+		if nid == "" || seen[nid] {
+			continue
+		}
+		seen[nid] = true
+		out = append(out, scrollPointToResult(pt.Payload, score))
+	}
+	return out
+}
+
+// scrollPointToResult maps a scroll-point payload to a LegalSearchResult. Mirrors
+// hitsToResults' payload keys; the score is assigned by the caller (concept rank).
+func scrollPointToResult(payload map[string]interface{}, score float64) LegalSearchResult {
+	regCode := getString(payload, "regulation_code")
+	if regCode == "" {
+		regCode = getString(payload, "regulation_id")
+	}
+	return LegalSearchResult{
+		Text:            getString(payload, "chunk_text"),
+		RegulationCode:  regCode,
+		RegulationName:  getString(payload, "regulation_name_de"),
+		RegulationShort: getString(payload, "regulation_short"),
+		Category:        getString(payload, "category"),
+		Article:         getString(payload, "article"),
+		ArticleLabel:    getString(payload, "article_label"),
+		Paragraph:       getString(payload, "paragraph"),
+		SourceURL:       getString(payload, "source_url"),
+		CitationUnit:    getString(payload, "citation_unit"),
+		Score:           score,
+	}
+}

ai-compliance-sdk/internal/ucca/concept_ontology.go

@@ -0,0 +1,97 @@
+package ucca
+
+import (
+	"sort"
+	"strings"
+)
+
+// Legal Concept Ontology — the fachliche IP bridge for the Concept->Norm recall
+// injector. The words users type ("Datenschutzerklärung", "Cookie Banner") are
+// rarely identical to the article titles that actually govern them (Art. 12/13/14
+// DSGVO, § 25 TDDDG). Embedding similarity misses this leap, so these bridges are
+// curated: concept keyword -> load-bearing norm_ids. This is NOT a fallback to
+// hardcoding — it is domain knowledge that surfaces the normatively load-bearing
+// units within the (already correctly retrieved) documents.
+type conceptNorm struct {
+	keywords []string
+	normIDs  []string
+}
+
+var legalConceptOntology = []conceptNorm{
+	{[]string{"datenschutzerklärung", "datenschutzerklaerung", "privacy policy", "datenschutzhinweise", "datenschutzinformation"},
+		[]string{"EU-DSGVO-Art12", "EU-DSGVO-Art13", "EU-DSGVO-Art14"}},
+	{[]string{"cookie banner", "cookie-banner", "cookies", "cookie", "tracking"},
+		[]string{"DE-TDDDG-§25", "EU-DSGVO-Art6", "EU-DSGVO-Art7"}},
+	{[]string{"dsfa", "folgenabschätzung", "folgenabschaetzung", "datenschutz-folgenabschätzung"},
+		[]string{"EU-DSGVO-Art35", "EU-DSGVO-Art36"}},
+	{[]string{"auskunft", "auskunftsrecht", "auskunftsersuchen"},
+		[]string{"EU-DSGVO-Art15"}},
+	{[]string{"löschung", "loeschung", "vergessenwerden", "recht auf vergessen"},
+		[]string{"EU-DSGVO-Art17"}},
+	{[]string{"datenübertragbarkeit", "datenuebertragbarkeit", "portabilität", "portabilitaet"},
+		[]string{"EU-DSGVO-Art20"}},
+	{[]string{"widerspruch", "widerspruchsrecht"},
+		[]string{"EU-DSGVO-Art21"}},
+	{[]string{"datenpanne", "datenschutzverletzung", "data breach", "verletzung des schutzes"},
+		[]string{"EU-DSGVO-Art33", "EU-DSGVO-Art34"}},
+	// E4-Quick-Curation (2026-07-01): resolved abbreviations (E2) pull their core norms.
+	{[]string{"technische und organisatorische maßnahmen", "technische und organisatorische massnahmen"},
+		[]string{"EU-DSGVO-Art32", "EU-DSGVO-Art25", "EU-DSGVO-Art5"}},
+	{[]string{"verzeichnis von verarbeitungstätigkeiten", "verzeichnis von verarbeitungstaetigkeiten", "verarbeitungsverzeichnis"},
+		[]string{"EU-DSGVO-Art30"}},
+	{[]string{"auftragsverarbeitungsvertrag", "auftragsverarbeitung", "auftragsverarbeiter"},
+		[]string{"EU-DSGVO-Art28"}},
+	{[]string{"datenschutzbeauftragt"},
+		[]string{"EU-DSGVO-Art37", "EU-DSGVO-Art38", "EU-DSGVO-Art39"}},
+}
+
+// ConceptNorms returns the load-bearing norm_ids for the concepts named in the
+// query (dedup, order-preserving). Empty if no concept is named.
+func ConceptNorms(query string) []string {
+	q := normalizeGerman(query)
+	seen := map[string]bool{}
+	out := []string{}
+	for _, cn := range legalConceptOntology {
+		for _, kw := range cn.keywords {
+			if strings.Contains(q, normalizeGerman(kw)) {
+				for _, nid := range cn.normIDs {
+					if !seen[nid] {
+						seen[nid] = true
+						out = append(out, nid)
+					}
+				}
+				break
+			}
+		}
+	}
+	return out
+}
+
+// InjectConceptNorms merges concept-injected norm units into the results so the
+// load-bearing norms are VISIBLE in the evidence set. Dedups by citation_unit
+// (skips norms already retrieved), then re-sorts by score — the injected units
+// carry a just-below-top score so they surface high WITHOUT displacing the top
+// document hit (inject, don't blindly dominate). Caps at topK.
+func InjectConceptNorms(results, injected []LegalSearchResult, topK int) []LegalSearchResult {
+	if len(injected) == 0 {
+		return results
+	}
+	present := map[string]bool{}
+	for _, r := range results {
+		if r.CitationUnit != "" {
+			present[r.CitationUnit] = true
+		}
+	}
+	merged := append([]LegalSearchResult{}, results...)
+	for _, in := range injected {
+		if in.CitationUnit != "" && !present[in.CitationUnit] {
+			merged = append(merged, in)
+			present[in.CitationUnit] = true
+		}
+	}
+	sort.SliceStable(merged, func(i, j int) bool { return merged[i].Score > merged[j].Score })
+	if topK > 0 && len(merged) > topK {
+		merged = merged[:topK]
+	}
+	return merged
+}

ai-compliance-sdk/internal/ucca/concept_test.go

@@ -0,0 +1,48 @@
+package ucca
+
+import "testing"
+
+func TestConceptNorms(t *testing.T) {
+	q := "Was muss ich beachten wenn ich meine Datenschutzerklärung schreibe für meine Website mit Cookie Banner?"
+	got := ConceptNorms(q)
+	want := map[string]bool{
+		"EU-DSGVO-Art12": true, "EU-DSGVO-Art13": true, "EU-DSGVO-Art14": true,
+		"DE-TDDDG-§25": true, "EU-DSGVO-Art6": true, "EU-DSGVO-Art7": true,
+	}
+	for _, nid := range got {
+		delete(want, nid)
+	}
+	if len(want) > 0 {
+		t.Errorf("ConceptNorms missing %v; got %v", want, got)
+	}
+	if len(ConceptNorms("Wie ist das Wetter heute?")) != 0 {
+		t.Errorf("no concept named should yield no norms")
+	}
+}
+
+func TestInjectConceptNorms(t *testing.T) {
+	results := []LegalSearchResult{
+		{CitationUnit: "DSK OH Telemedien", Score: 0.98},
+		{CitationUnit: "Art. 25 DSGVO", Score: 0.95},
+	}
+	injected := []LegalSearchResult{
+		{CitationUnit: "Art. 13 DSGVO", Score: 0.979},
+		{CitationUnit: "Art. 25 DSGVO", Score: 0.979}, // already present -> must not double
+	}
+	out := InjectConceptNorms(results, injected, 10)
+	if out[0].CitationUnit != "DSK OH Telemedien" {
+		t.Errorf("top document hit must stay #1 (not dominated), got %s", out[0].CitationUnit)
+	}
+	if len(out) != 3 {
+		t.Errorf("expected 3 (Art.25 not duplicated), got %d", len(out))
+	}
+	found := false
+	for _, r := range out {
+		if r.CitationUnit == "Art. 13 DSGVO" {
+			found = true
+		}
+	}
+	if !found {
+		t.Errorf("Art. 13 DSGVO must be injected + visible")
+	}
+}

ai-compliance-sdk/internal/ucca/embed_cache.go

@@ -0,0 +1,34 @@
+package ucca
+
+import "context"
+
+type embCacheKeyT struct{}
+
+var embCacheKey embCacheKeyT
+
+type embCacheEntry struct {
+	query string
+	vec   []float64
+}
+
+// embedForQuery returns the query embedding, reusing a value precomputed for the SAME
+// query and stashed in ctx by withQueryEmbedding. This collapses the Authority Router's
+// per-collection fan-out from N embeddings to ONE — decisive when the embedding endpoint
+// is remote (dev/OVH), where N round-trips dominated /retrieve latency. Falls back to a
+// fresh embedding when nothing is cached (direct Search / SearchCollection callers).
+func (c *LegalRAGClient) embedForQuery(ctx context.Context, query string) ([]float64, error) {
+	if v, ok := ctx.Value(embCacheKey).(*embCacheEntry); ok && v.query == query && len(v.vec) > 0 {
+		return v.vec, nil
+	}
+	return c.generateEmbedding(ctx, query)
+}
+
+// withQueryEmbedding precomputes the query embedding once and stashes it in ctx so the
+// concurrent per-collection searches reuse it instead of each re-embedding. Best-effort:
+// on embed error the ctx is returned unchanged and callers fall back to per-call embedding.
+func (c *LegalRAGClient) withQueryEmbedding(ctx context.Context, query string) context.Context {
+	if vec, err := c.generateEmbedding(ctx, query); err == nil && len(vec) > 0 {
+		return context.WithValue(ctx, embCacheKey, &embCacheEntry{query: query, vec: vec})
+	}
+	return ctx
+}

ai-compliance-sdk/internal/ucca/evidence_type.go

@@ -0,0 +1,68 @@
+package ucca
+
+import "context"
+
+// EvidenceType classifies a retrieved unit by WHAT KIND of evidence it is, independent of its
+// collection. Footnotes/tables/figures are Evidence Types, not collections. The Authority Router
+// surfaces non-text evidence from the authoritative knowledge space (the KB slice) SEPARATELY from
+// the merged text top-K, so fine-grained evidence isn't outranked by broad-base text.
+//
+// The layer this introduces: Intent -> Knowledge Space -> EvidenceType -> Collection -> Merge ->
+// Authority. Today FOOTNOTE is populated; FIGURE arrives with C8 and TABLE is already present from
+// C6/C9 — no router rebuild needed, the same path carries every new evidence type.
+type EvidenceType string
+
+const (
+	EvidenceText     EvidenceType = "text"
+	EvidenceFootnote EvidenceType = "footnote"
+	EvidenceTable    EvidenceType = "table"
+	EvidenceFigure   EvidenceType = "figure"
+)
+
+// classifyEvidence derives the EvidenceType from a result's payload markers. Precedence
+// footnote > figure > table > text (a unit carries at most one is_* marker in practice).
+func classifyEvidence(r LegalSearchResult) EvidenceType {
+	switch {
+	case r.IsFootnote:
+		return EvidenceFootnote
+	case r.IsFigure:
+		return EvidenceFigure
+	case r.IsTable:
+		return EvidenceTable
+	default:
+		return EvidenceText
+	}
+}
+
+// evidenceRetrievalTopK is the budget for the authoritative-KB evidence pass. Deliberately targeted
+// (the authoritative slice within the recognized knowledge space), NOT a blanket top-K increase of
+// the merged result set — the successes came from BETTER-targeted evidence, not MORE evidence.
+const evidenceRetrievalTopK = 20
+
+// maxEvidencePerType caps each surfaced evidence type.
+const maxEvidencePerType = 6
+
+// RetrieveEvidence returns the authoritative typed evidence (footnotes/tables/figures) for an
+// in-scope query, pulled from the KB slice and grouped by EvidenceType. This is the "Evidence Type"
+// router layer (Option A): when the query is in the KB knowledge space, the authoritative evidence
+// within that space is surfaced separately so it isn't lost in the broad-base text merge. Returns an
+// empty map when out of scope or KB routing is disabled. Text evidence is NOT returned here — it
+// flows through the normal Retrieve() merge (the LLM context + the sources list).
+func (c *LegalRAGClient) RetrieveEvidence(ctx context.Context, query string) map[EvidenceType][]LegalSearchResult {
+	ev := map[EvidenceType][]LegalSearchResult{}
+	if !c.kbScopeRoutingEnabled || c.kbSliceCollection == "" || !inKBScope(query) {
+		return ev
+	}
+	hits, err := c.searchInternal(ctx, c.kbSliceCollection, query, nil, evidenceRetrievalTopK)
+	if err != nil {
+		return ev
+	}
+	for _, h := range hits {
+		t := classifyEvidence(h)
+		if t == EvidenceText || len(ev[t]) >= maxEvidencePerType {
+			continue
+		}
+		ev[t] = append(ev[t], h)
+	}
+	return ev
+}

ai-compliance-sdk/internal/ucca/filter_test.go

@@ -0,0 +1,26 @@
+package ucca
+
+import "testing"
+
+func TestFilterByKnowledgeSpace(t *testing.T) {
+	results := []LegalSearchResult{
+		{CitationUnit: "Art. 13 DSGVO", RegulationCode: "DSGVO"},
+		{CitationUnit: "EU Mdr", RegulationCode: "EU MDR"},
+		{CitationUnit: "UStG § 14", RegulationCode: "USTG"},
+		{CitationUnit: "DSK OH Telemedien", RegulationCode: "DSK OH TELEMEDIEN"},
+		{CitationUnit: "eIDAS", RegulationCode: "EIDAS"},
+	}
+	out := FilterByKnowledgeSpace(results, "datenschutz", 10)
+	for _, r := range out {
+		if KnowledgeSpaceOf(r.RegulationCode) != "datenschutz" {
+			t.Errorf("off-domain leaked into scoped result: %s (%s)", r.CitationUnit, r.RegulationCode)
+		}
+	}
+	if len(out) != 2 { // Art. 13 DSGVO + DSK OH Telemedien
+		t.Errorf("expected 2 datenschutz hits, got %d", len(out))
+	}
+	// domain with no hits -> fall back to input (never strand the answer)
+	if len(FilterByKnowledgeSpace(results, "maschinen", 10)) != len(results) {
+		t.Errorf("no-hit domain should fall back to full input")
+	}
+}

ai-compliance-sdk/internal/ucca/intent.go

@@ -0,0 +1,46 @@
+package ucca
+
+import "strings"
+
+// DetectIntent classifies the INTERACTION INTENT of a query (Advisor Reasoning
+// Stack E3). The same norms answer very differently depending on the TASK the user
+// wants: "Was ist X?" (definition) vs "Wie schreibe ich X?" (anleitung) vs "Prüfe X"
+// (review). The SDK detects the intent deterministically and emits it; the FE picks
+// the answer FORM, so the LLM gets a precise assignment ("write an Anleitung over
+// this evidence") instead of guessing the format. Returns "" (neutral) when no
+// clear task is signalled. First tier of ~20-30 intent types.
+func DetectIntent(query string) string {
+	q := " " + normalizeGerman(query) + " "
+	has := func(subs ...string) bool {
+		for _, s := range subs {
+			if strings.Contains(q, normalizeGerman(s)) {
+				return true
+			}
+		}
+		return false
+	}
+	switch {
+	case has("prüfe", "prüf mein", "überprüfe", "überprüf", "review", "checke mein",
+		"ist mein", "ist meine", "ist unser", "ist unsere", "konform", "stimmt mein",
+		"bewerte mein", "analysiere mein"):
+		return "review"
+	case has("checkliste", "was muss ich alles", "was gehört alles", "was gehört in",
+		"welche punkte muss", "was brauche ich alles"):
+		return "checkliste"
+	case has("vergleich", "unterschied", "worin unterscheid", " vs ", " versus ",
+		"gegenüber", "im gegensatz"):
+		return "vergleich"
+	case has("wie schreibe", "wie erstelle", "wie erstell", "wie mache", "wie baue",
+		"wie setze ich", "wie gehe ich vor", "wie formuliere", "wie richte ich",
+		"anleitung", "schritt für schritt", "schritt-für-schritt", "erstelle mir",
+		"erstell mir", "generiere", "was muss ich beachten", "worauf muss ich achten"):
+		return "anleitung"
+	case has("welche risiken", "welche gefahren", "risikoanalyse", "welche bedrohungen"):
+		return "risikoanalyse"
+	case has("was ist", "was bedeutet", "was versteht man", "was sind", "definition",
+		"erkläre mir", "erklär mir", "was heißt", "was genau ist"):
+		return "definition"
+	default:
+		return ""
+	}
+}

ai-compliance-sdk/internal/ucca/intent_test.go

@@ -0,0 +1,22 @@
+package ucca
+
+import "testing"
+
+func TestDetectIntent(t *testing.T) {
+	cases := map[string]string{
+		"Was ist eine Datenschutzerklärung?":                 "definition",
+		"Wie schreibe ich eine Datenschutzerklärung?":        "anleitung",
+		"Was muss ich beachten wenn ich eine DSE schreibe?":  "anleitung",
+		"Prüfe meine Datenschutzerklärung.":                  "review",
+		"Ist meine Datenschutzerklärung konform?":            "review",
+		"Vergleiche DSGVO und BDSG.":                         "vergleich",
+		"Welche Risiken gibt es?":                            "risikoanalyse",
+		"Erstelle mir eine Checkliste für die DSFA.":         "checkliste",
+		"Wie ist das Wetter?":                                "",
+	}
+	for q, want := range cases {
+		if got := DetectIntent(q); got != want {
+			t.Errorf("DetectIntent(%q)=%q want %q", q, got, want)
+		}
+	}
+}

ai-compliance-sdk/internal/ucca/knowledge_space.go

@@ -0,0 +1,148 @@
+package ucca
+
+import "strings"
+
+// KnowledgeSpace is the CHIP-level knowledge domain used by the clarity gate's
+// concentration signal + the user-facing context chips. It is deliberately RICHER
+// than the 4 authority domains in authority.go (data_protection/cyber/ai/
+// product_safety), which drive the EU-primary/subsidiarity rerank. The clarity
+// gate must reflect the FULL corpus breadth (arbeitsschutz, arbeitsrecht,
+// wirtschaftsrecht, finanz, ...) so a broad query surfaces as broad. Kept separate
+// + additive so the tuned authority rerank stays untouched. Corpus-grounded from
+// the 463 real regulation codes (0.3% fall through to "sonstiges").
+
+// knowledgeSpaceExact matches short/ambiguous codes by EXACT string (substring
+// would misfire on 2-3 char codes like "OR"/"AO"/"BGB").
+var knowledgeSpaceExact = map[string]string{
+	"HGB": "wirtschaftsrecht", "BGB": "wirtschaftsrecht", "AO": "wirtschaftsrecht", "OR": "wirtschaftsrecht",
+	"ABGB": "wirtschaftsrecht", "UGB": "wirtschaftsrecht", "IFRS": "wirtschaftsrecht", "BAO": "wirtschaftsrecht",
+	"GMBHG": "wirtschaftsrecht", "AKTG": "wirtschaftsrecht", "INSO": "wirtschaftsrecht", "USTG": "wirtschaftsrecht",
+	"GOBD": "wirtschaftsrecht", "EGBGB": "wirtschaftsrecht", "GEWO": "wirtschaftsrecht", "URHG": "wirtschaftsrecht",
+	"DPF": "datenschutz", "TKG": "datenschutz", "TMG": "datenschutz", "DDG": "datenschutz", "DSG": "datenschutz",
+	"DSV": "datenschutz", "DSM": "datenschutz", "SCC": "datenschutz", "EPRIVACY": "datenschutz",
+	"SCHREMS II": "datenschutz", "CH_REVDSG": "datenschutz", "PLANET49": "datenschutz", "GOOGLE FONTS": "datenschutz",
+	"DSA": "digitale_dienste", "DMA": "digitale_dienste", "DGA": "digitale_dienste", "EHDS": "digitale_dienste",
+	"EIDAS": "digitale_dienste", "EIDAS 2.0": "digitale_dienste", "DATA ACT": "digitale_dienste",
+	"DATAACT": "digitale_dienste", "DIGITAL CONTENT": "digitale_dienste",
+	"MVO": "produktsicherheit", "MACHINERY": "produktsicherheit", "MASCHVO": "produktsicherheit",
+	"MASCHINENVO": "produktsicherheit", "GPSR": "produktsicherheit", "PID": "produktsicherheit",
+	"EAA": "produktsicherheit", "BFSG": "produktsicherheit", "ELEKTROG": "produktsicherheit",
+	"VERPACKG": "produktsicherheit", "BATTVO": "produktsicherheit", "BATTDG": "produktsicherheit", "EU MDR": "produktsicherheit",
+	"DORA": "finanz", "PSD2": "finanz", "MICA": "finanz", "AMLR": "finanz", "VAIT": "finanz", "BAIT": "finanz", "GWG": "finanz",
+	"UWG": "verbraucherschutz", "UCPD": "verbraucherschutz", "VSBG": "verbraucherschutz", "PANGV": "verbraucherschutz",
+	"DL-INFOV": "verbraucherschutz", "OMNIBUS": "verbraucherschutz", "UWG AT": "verbraucherschutz",
+	"PRODHAFTG": "verbraucherschutz", "PRODUKTHAFTUNGS-RL": "verbraucherschutz",
+	"ARG": "arbeitsrecht",
+}
+
+// KnowledgeSpaceLabel maps a knowledge-space id to a user-facing chip label.
+var KnowledgeSpaceLabel = map[string]string{
+	"datenschutz": "Datenschutz", "cyber": "Cybersecurity", "ki": "KI",
+	"produktsicherheit": "Produktsicherheit", "arbeitsschutz": "Arbeitsschutz",
+	"arbeitsrecht": "Arbeitsrecht", "wirtschaftsrecht": "Wirtschaftsrecht",
+	"finanz": "Finanzregulierung", "digitale_dienste": "Digitale Dienste",
+	"verbraucherschutz": "Verbraucherschutz", "lieferkette": "Lieferkette/Nachhaltigkeit",
+	"hinweisgeber": "Hinweisgeberschutz", "sonstiges": "Sonstiges",
+}
+
+// KnowledgeSpaceOf maps a regulation_code to a knowledge space. Robust to code
+// variants (MVO/MASCHVO/MASCHINENVO -> produktsicherheit; DSK SDM / SDM B51 ->
+// datenschutz). Returns "" for empty/untagged codes (not a knowledge space).
+func KnowledgeSpaceOf(code string) string {
+	c := strings.ToUpper(strings.TrimSpace(code))
+	if c == "" || c == "NONE" {
+		return ""
+	}
+	if d, ok := knowledgeSpaceExact[c]; ok {
+		return d
+	}
+	has := func(subs ...string) bool {
+		for _, s := range subs {
+			if strings.Contains(c, s) {
+				return true
+			}
+		}
+		return false
+	}
+	pre := func(subs ...string) bool {
+		for _, s := range subs {
+			if strings.HasPrefix(c, s) {
+				return true
+			}
+		}
+		return false
+	}
+	switch {
+	case pre("TRGS", "TRBS", "ASR", "OSHA") || has("ARBSCHG", "GEFAHRSTOFF"):
+		return "arbeitsschutz"
+	case has("AI ACT", "KI-VO", "KI VERORDNUNG", "GPAI", "AI RMF", "HLEG AI", "GENAI", "OECD AI", "AI PRINCIPLES", "OH KI", "KI BEHOERDEN", "KI SICHERHEIT", "POS KI"):
+		return "ki"
+	case pre("DSGVO", "BDSG", "TDDDG", "DSK", "EDPB", "WP24", "WP25", "WP26", "DSFA", "BFDI", "BAYLDA", "BAYLFB", "EDPS") || has("DATENSCHUTZ", "LOESCHKONZEPT", "LOESCHUNG", "VVT", "TELEMEDIEN", "EU US DPF", "BESCHAEFTIGTENDATEN"):
+		return "datenschutz"
+	case has("CRA", "NIS2", "NISG", "BSIG", "BSI-TR", "BSI_KRITIS", "KRITIS", "ENISA", "NIST", "OWASP", "EUCSA", "EUCC", "CISA", "CYCLONEDX", "SPDX", "SLSA", "OPENTELEMETRY", "CVSS", "SECURE BY DESIGN"):
+		return "cyber"
+	case has("MACHINERY", "MASCH", "BLUE GUIDE", "FDA HFE"):
+		return "produktsicherheit"
+	case has("LKSG", "CSDDD", "CSRD", "TAXONOMY"):
+		return "lieferkette"
+	case has("HINSCHG", "GESCHGEHG"):
+		return "hinweisgeber"
+	case pre("BAG ", "BAG_") || has("ARBVG", "AZG", "ARBZG", "BETRVG", "KSCHG", "MUSCHG", "AGG", "MILOG", "TZBFG", "NACHWG", "BURLG", "611A", "PAY TRANSPARENCY", "ANGG", "MUTTERSCHUTZ"):
+		return "arbeitsrecht"
+	case has("ECOMMERCE", "ECG", "MEDIENG", "VERBRAUCHERRECHTE", "DIGITAL CONTENT"):
+		return "verbraucherschutz"
+	case pre("EUGH", "BVERFG", "BVGE", "BGH", "OGH") || has("EU TAXONOMY"):
+		return "wirtschaftsrecht"
+	default:
+		return "sonstiges"
+	}
+}
+
+// ScopeResults implements G1 scope-gating: when the query names a regulation, its
+// knowledge space's hits LEAD the result set (the L2 answer + [n] citations are
+// built on this order, so scoped answers cite the named regulation instead of the
+// embedding-majority domain). Non-scoped hits backfill to keep topK. Stable within
+// each partition. Returns results unchanged when scope is "".
+func ScopeResults(results []LegalSearchResult, scope string, topK int) []LegalSearchResult {
+	if scope == "" {
+		return results
+	}
+	scoped := make([]LegalSearchResult, 0, len(results))
+	rest := make([]LegalSearchResult, 0, len(results))
+	for _, r := range results {
+		if KnowledgeSpaceOf(r.RegulationCode) == scope {
+			scoped = append(scoped, r)
+		} else {
+			rest = append(rest, r)
+		}
+	}
+	out := append(scoped, rest...)
+	if topK > 0 && len(out) > topK {
+		out = out[:topK]
+	}
+	return out
+}
+
+// FilterByKnowledgeSpace returns ONLY the results in the given knowledge space —
+// a HARD scope with no off-domain backfill. Used by E5 context scoping: when the
+// user explicitly chose a domain chip, off-domain regelwerke (MDR/UStG/eIDAS) must
+// not reappear in the evidence. Falls back to the input when the domain has no hits
+// (never strand the answer). Caps topK.
+func FilterByKnowledgeSpace(results []LegalSearchResult, scope string, topK int) []LegalSearchResult {
+	if scope == "" {
+		return results
+	}
+	out := make([]LegalSearchResult, 0, len(results))
+	for _, r := range results {
+		if KnowledgeSpaceOf(r.RegulationCode) == scope {
+			out = append(out, r)
+		}
+	}
+	if len(out) == 0 {
+		return results
+	}
+	if topK > 0 && len(out) > topK {
+		out = out[:topK]
+	}
+	return out
+}

ai-compliance-sdk/internal/ucca/legal_rag_client.go

@@ -105,7 +105,7 @@ func (c *LegalRAGClient) searchInternal(ctx context.Context, collection string,
 		}
 	}
 
-	embedding, err := c.generateEmbedding(ctx, query)
+	embedding, err := c.embedForQuery(ctx, query)
 	if err != nil {
 		return nil, fmt.Errorf("failed to generate embedding: %w", err)
 	}

ai-compliance-sdk/internal/ucca/legal_rag_types.go

@@ -37,6 +37,17 @@ type LegalSearchResult struct {
 	// Supersede-Status (status="superseded", use_for_primary=false) — Alt-Quelle,
 	// die fuer Default-Fragen demoted wird (nicht versteckt; fuer Historie auffindbar).
 	Superseded bool `json:"-"`
+
+	// Evidence-Type-Marker — intern (json:"-", kein Pro-Result-Contract-Change), aus dem
+	// Qdrant-Payload befuellt. classifyEvidence() leitet daraus den EvidenceType ab; der
+	// Router surfacet nicht-Text-Evidence (Fußnote/Tabelle/Figur) getrennt vom Text-Merge,
+	// damit feingranulare Evidence nicht von Breit-Basis-Text ueberrankt wird.
+	IsFootnote       bool   `json:"-"`
+	FootnoteLabel    string `json:"-"`
+	FootnoteVerbatim string `json:"-"`
+	RefCitationUnit  string `json:"-"`
+	IsTable          bool   `json:"-"` // C6/C9: is_table (liniiert + borderless)
+	IsFigure         bool   `json:"-"` // C8: is_figure (noch nicht befuellt bis C8)
 }
 
 // LegalAssessment is the auditable explanation layer over a ranked result set:

ai-compliance-sdk/internal/ucca/multi_regulation.go

@@ -195,6 +195,13 @@ func hitsToResults(hits []qdrantSearchHit) []LegalSearchResult {
 			ReferencesOut:   getStringSlice(hit.Payload, "references_out"),
 			ReferencesIn:    getStringSlice(hit.Payload, "references_in"),
 			Superseded:      getString(hit.Payload, "status") == "superseded",
+
+			IsFootnote:       getBool(hit.Payload, "is_footnote"),
+			FootnoteLabel:    getString(hit.Payload, "footnote_label"),
+			FootnoteVerbatim: getString(hit.Payload, "footnote_verbatim"),
+			RefCitationUnit:  getString(hit.Payload, "ref_citation_unit"),
+			IsTable:          getBool(hit.Payload, "is_table"),
+			IsFigure:         getBool(hit.Payload, "is_figure"),
 		}
 	}
 	return results

ai-compliance-sdk/internal/ucca/text_norm.go

@@ -0,0 +1,15 @@
+package ucca
+
+import "strings"
+
+// normalizeGerman lowercases and folds German umlauts / ß to their ASCII digraphs
+// (ä→ae, ö→oe, ü→ue, ß→ss) so keyword matching is insensitive to whether the user
+// typed "Prüfe" or "Pruefe", "Datenschutzerklärung" or "Datenschutzerklaerung".
+// Applied to BOTH the query and the keyword lists in the German-text matchers.
+func normalizeGerman(s string) string {
+	return umlautFolder.Replace(strings.ToLower(s))
+}
+
+var umlautFolder = strings.NewReplacer(
+	"ä", "ae", "ö", "oe", "ü", "ue", "ß", "ss",
+)

ai-compliance-sdk/internal/ucca/umlaut_test.go

@@ -0,0 +1,29 @@
+package ucca
+
+import "testing"
+
+func TestDetectIntentUmlautFold(t *testing.T) {
+	cases := map[string]string{
+		"Pruefe meine Datenschutzerklaerung.": "review",     // ASCII digraph
+		"Prüfe meine Datenschutzerklärung.":   "review",     // umlaut
+		"Ueberpruefe das Impressum":           "review",     // ASCII "überprüfe"
+		"Was ist eine TOM?":                   "definition", // unchanged
+	}
+	for q, want := range cases {
+		if got := DetectIntent(q); got != want {
+			t.Errorf("DetectIntent(%q)=%q want %q", q, got, want)
+		}
+	}
+}
+
+func TestConceptNormsUmlautFold(t *testing.T) {
+	// ASCII "datenschutzerklaerung" must resolve to the same core norms as the umlaut form.
+	ascii := ConceptNorms("Was gehoert in eine Datenschutzerklaerung?")
+	umlaut := ConceptNorms("Was gehört in eine Datenschutzerklärung?")
+	if len(ascii) == 0 {
+		t.Errorf("ConceptNorms(ASCII datenschutzerklaerung) returned none")
+	}
+	if len(ascii) != len(umlaut) {
+		t.Errorf("ASCII vs umlaut concept norms differ: %v vs %v", ascii, umlaut)
+	}
+}

docs-src/development/obligation_registry_v1.md

@@ -128,3 +128,74 @@ SBOM ✓  →  Vuln ✓  →  Registry v1 (DIESE Spec)  →  Ontologie/Beziehung
 Begründung: Schema jetzt billig änderbar; bei 300–1000 Obligations wird jede Schemaänderung
 teuer. Fortschritt wird daran gemessen, ob jede neue Obligation die Registry besser macht —
 nicht an neuen Controls.
+
+## Scope-Audit (Review-Step, PFLICHT je Cut)
+
+Die Registry modelliert **Hersteller-Pflichten**. Bestimmungen, die an Behörden / notifizierte
+Stellen / Mitgliedstaaten adressiert sind (Sanktionen, Marktüberwachung, Anforderungen an
+Konformitätsbewertungsstellen), sind Enforcement-/Institutions-Recht. **Prinzip: Adressat der Norm
+⊥ Handlungspflicht des Herstellers.** `scope`-Attribut-Achse (Enum, KEINE neue Objektklasse):
+
+- `in_scope` — Norm adressiert direkt den Hersteller (Default).
+- `out_of_scope` — reines Staats-/Durchsetzungs-/Institutions-Recht (Adressat ≠ Hersteller, KEINE
+  mittelbare Herstellerpflicht). Aus `obligation_join_keys.json` gefiltert. Präzedenz CSIRT/ENISA.
+- `derived_obligation` — Norm adressiert primär eine andere Rolle, erzeugt aber MITTELBAR eine
+  Hersteller-Handlungspflicht → **bleibt im Set** (`scope_split_candidate` markiert spätere
+  Aufspaltung Normadressat ↔ abgeleitete Pflicht; nicht vorzeitig festziehen).
+
+**Gate-Regel:**
+```
+Jeder neue Obligation-Cut muss durch Scope-Audit laufen.
+Findings mit authority-/institution-addressed obligations werden dokumentiert.
+Automatische Reclassification ist verboten, solange kein explizites Review-Go vorliegt.
+```
+
+**Werkzeug-Trennung (FLAG ⊥ MUTATE):**
+- `scope_audit.py` — **flaggt nur** (scannt alle Registries → `scope_audit_findings.json`; mutiert nie).
+- `validate_registry.py` — surfaced pro Cut unklassifizierte authority-/institution-Obligations als
+  **non-fatal Warnung** (blockt nicht, mutiert nicht).
+- `apply_scope_classification.py` — **mutiert** (setzt `scope`), läuft NUR nach explizitem Review-Go
+  (ändert `join_keys` + Compliance-Execution-Sync → menschlich/koordiniert).
+
+## Weg 1 — Obligation→Norm-Zitierfähigkeit (Datenbereitschaft; UI deferred)
+
+**Zwei Zitierebenen — NICHT verwechseln (beide langfristig nötig):**
+- **RAG-Evidence-Zitat** (heute im Advisor sichtbar): `Frage/Antwort → Evidence-Chunk`
+  (KB-v2 `article_label`/`source_url`, `[n]`-Citations). Quelle = Compliance/KB-Session `/retrieve`.
+- **Obligation→Norm-Zitat** (der `norm_id`-Join): `Pflicht → konkrete Rechtsgrundlage` (KB-v2-Unit).
+  Quelle = diese Registry. **Weg 1** = dieses Zitat sichtbar machen.
+
+**Datenbereitschaft (Stand 2026-07-01):**
+
+| Baustein | Status |
+|---|---|
+| `obligations/*.json` → `legal_basis.norm_ids` | ✅ 62/64 joinbar (53 annex + 16 article, KB-v2-verifiziert) |
+| KB-v2-Join-Ziele (CRA Art1-71 · MaschVO Art1-54 · Annexe) | ✅ bestätigt |
+| `obligation-status`-Endpoint + Traversal (`obligation_id→citation_unit→Controls→Evidence`) | ✅ vorhanden; exponiert `LegalBasis = citation_units` |
+| Runtime-Vertrag `obligation_join_keys.json` trägt `norm_ids` | ❌ nur `citation_units` (Anker-Strings), KEINE `norm_ids` |
+| Go `ObligationKey`-Struct / `CitationSpans` | ❌ kein `NormIDs`-Feld; `AssessObligationStatus` setzt `CitationSpans:"pending"` hart (compliance_status.go) |
+
+**Deferred-Sequenz (UI zuletzt) — bewusst NICHT gebaut:**
+1. **Data-Prep** (nicht-UI, Domäne 2): `norm_ids` in `export_join_keys.py` → `obligation_join_keys.json` exportieren.
+2. **Build** (ai-sdk, koordiniert): `NormIDs` in `ObligationKey`; `AssessObligationStatus` füllt `CitationSpans` aus `norm_ids` statt `"pending"`.
+3. **UI** (später): „Diese Pflicht beruht auf **CRA Anhang I / Art. 13**".
+
+**Design-Vorgabe für Phase B (Runtime-Vertrag): generische Legal-Reference-Hülle statt flachem
+`norm_ids`.** `norm_ids` sind nur die ERSTE Rechtsreferenz-Art. Damit der Runtime-Vertrag
+(`obligation_join_keys.json` + Go `ObligationKey`) bei neuen Referenz-Arten NICHT erneut geändert
+werden muss, trägt jeder Eintrag eine **erweiterbare Hülle** (optionale Keys, additiv):
+```json
+"legal_reference": {          // Hülle — NICHT "legal_basis" (kollidiert mit dem Obligation-Array)
+  "norm_ids":        [...],   // bindendes Primärrecht (Artikel/Anhang)  ← Phase B baut nur DAS
+  "citation_units":  [...],   // menschliche Anker-Strings (Interim-Brücke)
+  "recital_ids":     [...],   // Erwägungsgründe        (später, additiv)
+  "guidance_ids":    [...],   // Leitlinien EDPB/DSK    (später, additiv)
+  "case_law_ids":    [...],   // Gerichtsurteile        (später, additiv)
+  "interpretation_ids": [...] // Interpretationshilfen  (später, additiv)
+}
+```
+Prinzip: neue Referenz-**Art** = neuer optionaler Key in der Hülle → Vertrag/Go-Struct bleiben
+stabil (wie die `scope`-Achse: erweitern über Attribute, nicht über Strukturumbau). **Bindend ⊥
+Guidance-Trennung bleibt erhalten** (`norm_ids`/`case_law` = bindend · `guidance_ids` = Soft-Law) —
+konsistent mit `legal_basis` ⊥ `guidance_basis` und dem Authority-Router. In Phase B wird NUR
+`norm_ids` befüllt; die übrigen Keys sind reservierter Platz, kein Bauauftrag.

obligations/cra.json

@@ -23,7 +23,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (1)",
-     "citation": "SBOM in gängigem maschinenlesbarem Format, mind. Top-Level-Abhängigkeiten"
+     "citation": "SBOM in gängigem maschinenlesbarem Format, mind. Top-Level-Abhängigkeiten",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -128,7 +132,7 @@
    "member_count": 85,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -149,7 +153,12 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Art. 3(36) i.V.m. Annex I Part II (1)",
-     "citation": "SBOM-Definition: formale Aufzeichnung enthaltener Komponenten und Abhängigkeiten"
+     "citation": "SBOM-Definition: formale Aufzeichnung enthaltener Komponenten und Abhängigkeiten",
+     "norm_ids": [
+      "EU-CRA-AnhangI",
+      "EU-CRA-Art3"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -188,7 +197,7 @@
    "member_count": 24,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -209,7 +218,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (1)",
-     "citation": "gängiges, maschinenlesbares Format"
+     "citation": "gängiges, maschinenlesbares Format",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -248,7 +261,7 @@
    "member_count": 19,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -269,7 +282,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (1)",
-     "citation": "SBOM während Support-Zeitraum führen"
+     "citation": "SBOM während Support-Zeitraum führen",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -315,7 +332,7 @@
    "member_count": 31,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -476,7 +493,12 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Art. 31 / Annex I Part II (1)",
-     "citation": "Vorlage der SBOM auf begründetes Verlangen der Marktüberwachungsbehörde"
+     "citation": "Vorlage der SBOM auf begründetes Verlangen der Marktüberwachungsbehörde",
+     "norm_ids": [
+      "EU-CRA-AnhangI",
+      "EU-CRA-Art31"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -493,7 +515,7 @@
    "member_count": 8,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -514,7 +536,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Art. 31(4)",
-     "citation": "Marktüberwachungsbehörden wahren Vertraulichkeit der erhaltenen Informationen"
+     "citation": "Marktüberwachungsbehörden wahren Vertraulichkeit der erhaltenen Informationen",
+     "norm_ids": [
+      "EU-CRA-Art31"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [
@@ -539,7 +565,7 @@
    "member_count": 10,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -600,7 +626,12 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Art. 31 i.V.m. Annex VII",
-     "citation": "technische Dokumentation muss SBOM-relevante Nachweise enthalten"
+     "citation": "technische Dokumentation muss SBOM-relevante Nachweise enthalten",
+     "norm_ids": [
+      "EU-CRA-AnhangVII",
+      "EU-CRA-Art31"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -628,7 +659,7 @@
    "member_count": 13,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -649,7 +680,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (1)",
-     "citation": "Komponenten identifizieren und dokumentieren, einschl. SBOM"
+     "citation": "Komponenten identifizieren und dokumentieren, einschl. SBOM",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -717,7 +752,7 @@
    "member_count": 48,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -738,7 +773,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (1)",
-     "citation": "Schwachstellen behandeln und beheben"
+     "citation": "Schwachstellen behandeln und beheben",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -819,7 +858,7 @@
    "member_count": 61,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -840,7 +879,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (2) & (8)",
-     "citation": "Schwachstellen unverzüglich beheben, kostenlose Sicherheitsupdates"
+     "citation": "Schwachstellen unverzüglich beheben, kostenlose Sicherheitsupdates",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -965,7 +1008,7 @@
    "member_count": 110,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "merged_into": "provide_security_updates",
    "status": "deprecated_alias",
@@ -989,7 +1032,12 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Article 13(8) & Annex VII",
-     "citation": "Schwachstellenbehandlungsprozesse einrichten und in technischer Doku belegen"
+     "citation": "Schwachstellenbehandlungsprozesse einrichten und in technischer Doku belegen",
+     "norm_ids": [
+      "EU-CRA-AnhangVII",
+      "EU-CRA-Art13"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1114,7 +1162,7 @@
    "member_count": 105,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -1135,7 +1183,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (5)",
-     "citation": "Coordinated Vulnerability Disclosure Policy einrichten"
+     "citation": "Coordinated Vulnerability Disclosure Policy einrichten",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1233,7 +1285,7 @@
    "member_count": 78,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -1254,7 +1306,12 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Article 14 & Article 16",
-     "citation": "Meldepflicht aktiv ausgenutzter Schwachstellen über Single Reporting Platform"
+     "citation": "Meldepflicht aktiv ausgenutzter Schwachstellen über Single Reporting Platform",
+     "norm_ids": [
+      "EU-CRA-Art14",
+      "EU-CRA-Art16"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1294,7 +1351,7 @@
    "member_count": 31,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   },
   {
@@ -1315,7 +1372,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I Part II (4) & (6)",
-     "citation": "Informationen über behobene Schwachstellen teilen und offenlegen"
+     "citation": "Informationen über behobene Schwachstellen teilen und offenlegen",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1335,7 +1396,7 @@
    "member_count": 5,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft"
   }
  ],
@@ -1581,5 +1642,13 @@
   "produces_evidence_for",
   "implements",
   "derived_from"
- ]
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/cra_authentication.json

@@ -26,7 +26,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(d)",
-     "citation": "protect... by ensuring protection from unauthorised access, including by reporting... appropriate control mechanisms incl. authentication, identity or access management"
+     "citation": "protect... by ensuring protection from unauthorised access, including by reporting... appropriate control mechanisms incl. authentication, identity or access management",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1391,7 +1395,7 @@
    "member_count": 1339,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.95,
@@ -4682,7 +4686,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(e)",
-     "citation": "protect the confidentiality... through state-of-the-art mechanisms incl. encryption"
+     "citation": "protect the confidentiality... through state-of-the-art mechanisms incl. encryption",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -5277,7 +5285,7 @@
    "member_count": 533,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -5650,7 +5658,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(e)",
-     "citation": "protect the confidentiality of stored, transmitted or otherwise processed data"
+     "citation": "protect the confidentiality of stored, transmitted or otherwise processed data",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -5994,7 +6006,7 @@
    "member_count": 315,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -6326,7 +6338,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(a)",
-     "citation": "be made available with a secure by default configuration"
+     "citation": "be made available with a secure by default configuration",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -6347,7 +6363,7 @@
    "member_count": 9,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -8408,7 +8424,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(e)",
-     "citation": "protect the confidentiality of... transmitted... data... incl. encryption in transit"
+     "citation": "protect the confidentiality of... transmitted... data... incl. encryption in transit",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -8485,7 +8505,7 @@
    "member_count": 57,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -10214,7 +10234,11 @@
      "source": "CRA",
      "regulation_code": "eu_2024_2847",
      "anchor": "Annex I (2)(c)",
-     "citation": "ensure that vulnerabilities can be addressed through security updates... ensuring integrity"
+     "citation": "ensure that vulnerabilities can be addressed through security updates... ensuring integrity",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -10273,7 +10297,7 @@
    "member_count": 37,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.78,
@@ -10458,5 +10482,13 @@
   ],
   "from_obligations": 54,
   "to_obligations": 29
+ },
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
  }
 }

obligations/cra_core.json

@@ -23,7 +23,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(j)",
-     "citation": "limit attack surfaces, including external interfaces"
+     "citation": "limit attack surfaces, including external interfaces",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -38,7 +42,7 @@
     "component_remote_interface_security"
    ],
    "primary_implementation": "NIST CM-7",
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "core_from_5b"
   },
   {
@@ -56,7 +60,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(f)",
-     "citation": "protect the integrity of stored, transmitted or processed data, software and configuration"
+     "citation": "protect the integrity of stored, transmitted or processed data, software and configuration",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -74,9 +82,17 @@
     "code_signing"
    ],
    "primary_implementation": "NIST SI-7",
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "core_from_5b"
   }
  ],
- "relationships": []
+ "relationships": [],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/cra_logging.json

@@ -44,7 +44,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "monitor relevant internal activity, including the access to or modification of data, services or functions, where applicable, through recording and monitoring"
+     "citation": "monitor relevant internal activity, including the access to or modification of data, services or functions, where applicable, through recording and monitoring",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1038,7 +1042,7 @@
    "member_count": 961,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.95,
@@ -1066,7 +1070,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "recording and monitoring access to or modification of data, services or functions"
+     "citation": "recording and monitoring access to or modification of data, services or functions",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1601,7 +1609,7 @@
    "member_count": 505,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.92,
@@ -1629,7 +1637,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "monitor relevant internal activity including access to or modification of functions"
+     "citation": "monitor relevant internal activity including access to or modification of functions",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1878,7 +1890,7 @@
    "member_count": 226,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1906,7 +1918,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "recording and monitoring ... in a secure manner"
+     "citation": "recording and monitoring ... in a secure manner",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -2442,7 +2458,7 @@
    "member_count": 505,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.93,
@@ -2470,7 +2486,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "in a secure manner"
+     "citation": "in a secure manner",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -2550,7 +2570,7 @@
    "member_count": 59,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.88,
@@ -2942,7 +2962,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I Part I (2)(k)",
-     "citation": "monitor relevant internal activity"
+     "citation": "monitor relevant internal activity",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -3251,7 +3275,7 @@
    "member_count": 283,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -4737,5 +4761,13 @@
    ],
    "note": "M8/M5/M81 KI-/FRT- bzw. domaenenspezifische Trainings-/PIN-/Biometrie-Protokollierung (AI Act/sektorale Regulierung); M58/M59/M71/M56/M63 reine DSGVO-/datenschutzrechtliche bzw. nationale Verwaltungs-Protokollierungspflichten, nicht CRA Annex I (2)(k)"
   }
- ]
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/cra_machinery.json

@@ -48,7 +48,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1 (Allgemeine Grundsätze)",
-     "citation": "Der Hersteller einer Maschine hat eine Risikobeurteilung durchzuführen, um die für die Maschine geltenden Sicherheits- und Gesundheitsschutzanforderungen zu ermitteln."
+     "citation": "Der Hersteller einer Maschine hat eine Risikobeurteilung durchzuführen, um die für die Maschine geltenden Sicherheits- und Gesundheitsschutzanforderungen zu ermitteln.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -560,7 +564,7 @@
    "member_count": 480,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.95,
@@ -588,7 +592,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang IV (Technische Unterlagen)",
-     "citation": "Die technischen Unterlagen müssen die Risikobeurteilung mit den Ergebnissen enthalten."
+     "citation": "Die technischen Unterlagen müssen die Risikobeurteilung mit den Ergebnissen enthalten.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIV"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -882,7 +890,7 @@
    "member_count": 278,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1268,7 +1276,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.1.2 (Grundsätze für die Integration der Sicherheit)",
-     "citation": "Verbleibende Restrisiken sind in der Betriebsanleitung anzugeben."
+     "citation": "Verbleibende Restrisiken sind in der Betriebsanleitung anzugeben.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1439,7 +1451,7 @@
    "member_count": 158,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1467,7 +1479,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.4 (Anforderungen an Schutzeinrichtungen)",
-     "citation": "Bewegliche Teile der Maschine sind so zu gestalten und zu bauen, dass jegliches Unfallrisiko durch Kontakt verhütet wird; trennende oder nichttrennende Schutzeinrichtungen sind vorzusehen."
+     "citation": "Bewegliche Teile der Maschine sind so zu gestalten und zu bauen, dass jegliches Unfallrisiko durch Kontakt verhütet wird; trennende oder nichttrennende Schutzeinrichtungen sind vorzusehen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -2024,7 +2040,7 @@
    "member_count": 530,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -2052,7 +2068,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.2.4 (Stillsetzen, Not-Halt)",
-     "citation": "Jede Maschine muss mit einer oder mehreren Notvorrichtungen ausgerüstet sein, mit denen sich drohende oder eintretende Gefahren abwenden lassen."
+     "citation": "Jede Maschine muss mit einer oder mehreren Notvorrichtungen ausgerüstet sein, mit denen sich drohende oder eintretende Gefahren abwenden lassen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -2101,7 +2121,7 @@
    "member_count": 32,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.88,
@@ -2129,7 +2149,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.2.1 (Sicherheit und Zuverlässigkeit von Steuerungen)",
-     "citation": "Steuerungen sind so zu gestalten, dass sie sicher und zuverlässig sind und Gefährdungssituationen verhindern."
+     "citation": "Steuerungen sind so zu gestalten, dass sie sicher und zuverlässig sind und Gefährdungssituationen verhindern.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -2367,7 +2391,7 @@
    "member_count": 214,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -2395,7 +2419,12 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang I (Liste der Sicherheitsbauteile), Art. 5",
-     "citation": "Sicherheitsbauteile gemäß Anhang I unterliegen den Anforderungen der Verordnung und der Konformitätsbewertung."
+     "citation": "Sicherheitsbauteile gemäß Anhang I unterliegen den Anforderungen der Verordnung und der Konformitätsbewertung.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangI",
+      "EU-MaschVO-Art5"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -2454,7 +2483,7 @@
    "member_count": 43,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -2482,7 +2511,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.7.4 (Betriebsanleitung)",
-     "citation": "Jeder Maschine muss eine Betriebsanleitung in der/den Amtssprache(n) des Mitgliedstaats beiliegen, in dem die Maschine in Verkehr gebracht wird."
+     "citation": "Jeder Maschine muss eine Betriebsanleitung in der/den Amtssprache(n) des Mitgliedstaats beiliegen, in dem die Maschine in Verkehr gebracht wird.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -2828,7 +2861,7 @@
    "member_count": 325,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.92,
@@ -2856,7 +2889,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.3.7/1.7.4",
-     "citation": "Gefährdungen durch Blockierung beweglicher Teile sind zu berücksichtigen und Maßnahmen zur sicheren Beseitigung in der Betriebsanleitung anzugeben."
+     "citation": "Gefährdungen durch Blockierung beweglicher Teile sind zu berücksichtigen und Maßnahmen zur sicheren Beseitigung in der Betriebsanleitung anzugeben.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3204,7 +3241,7 @@
    "member_count": 334,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -3232,7 +3269,11 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 25 (Konformitätsbewertungsverfahren)",
-     "citation": "Vor dem Inverkehrbringen führt der Hersteller das anwendbare Konformitätsbewertungsverfahren durch."
+     "citation": "Vor dem Inverkehrbringen führt der Hersteller das anwendbare Konformitätsbewertungsverfahren durch.",
+     "norm_ids": [
+      "EU-MaschVO-Art25"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3422,7 +3463,7 @@
    "member_count": 172,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.88,
@@ -3450,7 +3491,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang IV (Technische Unterlagen)",
-     "citation": "Die technischen Unterlagen müssen die Konstruktions-, Herstellungs- und Funktionsbeschreibung sowie die Risikobeurteilung enthalten."
+     "citation": "Die technischen Unterlagen müssen die Konstruktions-, Herstellungs- und Funktionsbeschreibung sowie die Risikobeurteilung enthalten.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIV"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3523,7 +3568,7 @@
    "member_count": 57,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -3551,7 +3596,12 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 21, Art. 22 (EU-Konformitätserklärung, CE-Kennzeichnung)",
-     "citation": "Der Hersteller stellt eine EU-Konformitätserklärung aus und bringt die CE-Kennzeichnung an."
+     "citation": "Der Hersteller stellt eine EU-Konformitätserklärung aus und bringt die CE-Kennzeichnung an.",
+     "norm_ids": [
+      "EU-MaschVO-Art21",
+      "EU-MaschVO-Art22"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3624,7 +3674,7 @@
    "member_count": 57,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -3652,7 +3702,12 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 10, Art. 11 (Pflichten der Hersteller)",
-     "citation": "Die Hersteller gewährleisten, dass ihre Maschinen gemäß den grundlegenden Sicherheits- und Gesundheitsschutzanforderungen konstruiert und hergestellt wurden."
+     "citation": "Die Hersteller gewährleisten, dass ihre Maschinen gemäß den grundlegenden Sicherheits- und Gesundheitsschutzanforderungen konstruiert und hergestellt wurden.",
+     "norm_ids": [
+      "EU-MaschVO-Art10",
+      "EU-MaschVO-Art11"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3723,7 +3778,7 @@
    "member_count": 55,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -3751,7 +3806,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III (Grundlegende Sicherheits- und Gesundheitsschutzanforderungen)",
-     "citation": "Maschinen müssen die in Anhang III aufgeführten grundlegenden Sicherheits- und Gesundheitsschutzanforderungen erfüllen."
+     "citation": "Maschinen müssen die in Anhang III aufgeführten grundlegenden Sicherheits- und Gesundheitsschutzanforderungen erfüllen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -3943,7 +4002,7 @@
    "member_count": 171,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -4029,7 +4088,12 @@
     {
      "source": "MaschVO",
      "anchor": "Kapitel IV (Notifizierung von Konformitätsbewertungsstellen)",
-     "citation": "Notifizierte Stellen müssen die Anforderungen an Unabhängigkeit, Kompetenz und Unparteilichkeit erfüllen."
+     "citation": "Notifizierte Stellen müssen die Anforderungen an Unabhängigkeit, Kompetenz und Unparteilichkeit erfüllen.",
+     "norm_ids": [
+      "EU-MaschVO-KapitelIV"
+     ],
+     "norm_id_status": "chapter_no_kb_unit",
+     "norm_id_note": "Kapitel-Ebene nicht als KB-v2-Unit gemintet (Compiler = Artikel+Annex). Re-Anchor auf Konstituenten-Artikel = Enhancement (KB-v2 hat die Artikel); NICHT geraten."
     }
    ],
    "guidance_basis": [],
@@ -4052,7 +4116,7 @@
    "member_count": 11,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "chapter_reanchor_pending",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -4061,7 +4125,11 @@
     "llm_model": "claude-opus-4-8",
     "synthesis_version": "v1"
    },
-   "family": "machinery"
+   "family": "machinery",
+   "scope": "derived_obligation",
+   "scope_reason": "Norm adressiert primär die notifizierte Stelle (Unabhängigkeit/Kompetenz/Unparteilichkeit), erzeugt aber mittelbare Hersteller-Pflichten: notifizierte Stelle einbeziehen, erforderliche Unterlagen bereitstellen, Konformitätsbewertung korrekt durchführen.",
+   "scope_split_candidate": true,
+   "scope_split_note": "Kandidat für spätere Aufspaltung: 'Normadressat' (Anforderungen AN die notifizierte Stelle = institutional/out_of_scope) ↔ 'abgeleitete Herstellerpflicht' (NB einbeziehen + Unterlagen + Konformitätsbewertung = in_scope). NICHT vorzeitig festziehen."
   },
   {
    "id": "market_surveillance_safeguard",
@@ -4080,7 +4148,13 @@
     {
      "source": "MaschVO",
      "anchor": "Kapitel V/VI (Marktüberwachung, Schutzklauselverfahren)",
-     "citation": "Mitgliedstaaten ergreifen geeignete Maßnahmen gegen Maschinen, die ein Risiko darstellen; die Kommission koordiniert Schutzmaßnahmen."
+     "citation": "Mitgliedstaaten ergreifen geeignete Maßnahmen gegen Maschinen, die ein Risiko darstellen; die Kommission koordiniert Schutzmaßnahmen.",
+     "norm_ids": [
+      "EU-MaschVO-KapitelV",
+      "EU-MaschVO-KapitelVI"
+     ],
+     "norm_id_status": "chapter_no_kb_unit",
+     "norm_id_note": "Kapitel-Ebene nicht als KB-v2-Unit gemintet (Compiler = Artikel+Annex). Re-Anchor auf Konstituenten-Artikel = Enhancement (KB-v2 hat die Artikel); NICHT geraten."
     }
    ],
    "guidance_basis": [],
@@ -4125,7 +4199,7 @@
    "member_count": 30,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "chapter_reanchor_pending",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -4134,7 +4208,9 @@
     "llm_model": "claude-opus-4-8",
     "synthesis_version": "v1"
    },
-   "family": "machinery"
+   "family": "machinery",
+   "scope": "out_of_scope",
+   "scope_reason": "Adressat = Marktüberwachungsbehörden/Kommission (Schutzmaßnahmen, Schutzklauselverfahren); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA."
   },
   {
    "id": "sanctions",
@@ -4153,7 +4229,11 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 50 (Sanktionen)",
-     "citation": "Die Mitgliedstaaten legen Vorschriften über Sanktionen für Verstöße gegen diese Verordnung fest."
+     "citation": "Die Mitgliedstaaten legen Vorschriften über Sanktionen für Verstöße gegen diese Verordnung fest.",
+     "norm_ids": [
+      "EU-MaschVO-Art50"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -4185,7 +4265,7 @@
    "member_count": 19,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -4194,7 +4274,9 @@
     "llm_model": "claude-opus-4-8",
     "synthesis_version": "v1"
    },
-   "family": "machinery"
+   "family": "machinery",
+   "scope": "out_of_scope",
+   "scope_reason": "Adressat = Mitgliedstaaten (legen Sanktionen fest); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA (CRA-Vuln-Cut)."
   },
   {
    "id": "scope_transition_application",
@@ -4213,7 +4295,13 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 1, Art. 53, Art. 54 (Anwendungsbereich, Übergangsbestimmungen, Geltungsbeginn)",
-     "citation": "Diese Verordnung gilt ab dem festgelegten Datum unmittelbar in allen Mitgliedstaaten; Übergangsbestimmungen regeln die Anwendbarkeit."
+     "citation": "Diese Verordnung gilt ab dem festgelegten Datum unmittelbar in allen Mitgliedstaaten; Übergangsbestimmungen regeln die Anwendbarkeit.",
+     "norm_ids": [
+      "EU-MaschVO-Art1",
+      "EU-MaschVO-Art53",
+      "EU-MaschVO-Art54"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -4316,7 +4404,7 @@
    "member_count": 85,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -4344,7 +4432,11 @@
     {
      "source": "MaschVO",
      "anchor": "Art. 18 (wesentliche Veränderung)",
-     "citation": "Wer eine wesentliche Veränderung an einer Maschine vornimmt, gilt als Hersteller und muss die Anforderungen erfüllen."
+     "citation": "Wer eine wesentliche Veränderung an einer Maschine vornimmt, gilt als Hersteller und muss die Anforderungen erfüllen.",
+     "norm_ids": [
+      "EU-MaschVO-Art18"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -4417,7 +4509,7 @@
    "member_count": 60,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -4445,7 +4537,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.1.9 (Schutz gegen Korrumpierung)",
-     "citation": "Die Maschine ist so zu konstruieren, dass die Verbindung mit anderen Geräten nicht zu einer gefährlichen Situation führt; Hard- und Software, die für sicherheitsrelevante Funktionen kritisch sind, sind gegen unbeabsichtigte oder vorsätzliche Korrumpierung zu schützen."
+     "citation": "Die Maschine ist so zu konstruieren, dass die Verbindung mit anderen Geräten nicht zu einer gefährlichen Situation führt; Hard- und Software, die für sicherheitsrelevante Funktionen kritisch sind, sind gegen unbeabsichtigte oder vorsätzliche Korrumpierung zu schützen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -4562,7 +4658,7 @@
    "member_count": 86,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -4590,7 +4686,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.1.9, Nr. 1.2.1",
-     "citation": "Sicherheitsrelevante Hard- und Software ist gegen unbeabsichtigte oder vorsätzliche Korrumpierung zu schützen; eine Korrumpierung darf nicht zu gefährlichen Situationen führen."
+     "citation": "Sicherheitsrelevante Hard- und Software ist gegen unbeabsichtigte oder vorsätzliche Korrumpierung zu schützen; eine Korrumpierung darf nicht zu gefährlichen Situationen führen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -4659,7 +4759,7 @@
    "member_count": 46,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -4687,7 +4787,12 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang I Teil A, Anhang III Nr. 1.2.1",
-     "citation": "Maschinen mit sich vollständig oder teilweise selbst entwickelndem Verhalten durch maschinelles Lernen gelten als Hochrisikomaschinen und unterliegen besonderen Anforderungen."
+     "citation": "Maschinen mit sich vollständig oder teilweise selbst entwickelndem Verhalten durch maschinelles Lernen gelten als Hochrisikomaschinen und unterliegen besonderen Anforderungen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangI",
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -4798,7 +4903,7 @@
    "member_count": 96,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -4826,7 +4931,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 3 (Mobile Maschinen) / Nr. 6",
-     "citation": "Mobile Maschinen sind so zu konstruieren, dass Risiken im Gefahrenbereich und bei Fernsteuerung beherrscht werden."
+     "citation": "Mobile Maschinen sind so zu konstruieren, dass Risiken im Gefahrenbereich und bei Fernsteuerung beherrscht werden.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -4866,7 +4975,7 @@
    "member_count": 23,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.78,
@@ -4894,7 +5003,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 2-6 (besondere Maschinenkategorien)",
-     "citation": "Für bestimmte Maschinenkategorien gelten zusätzliche grundlegende Sicherheitsanforderungen."
+     "citation": "Für bestimmte Maschinenkategorien gelten zusätzliche grundlegende Sicherheitsanforderungen.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -5026,7 +5139,7 @@
    "member_count": 111,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.75,
@@ -5054,7 +5167,11 @@
     {
      "source": "MaschVO",
      "anchor": "Anhang III Nr. 1.5.8/1.5.9, Nr. 1.7.4.2",
-     "citation": "Die Betriebsanleitung muss Angaben zu Luftschallemissionen und Vibrationen enthalten."
+     "citation": "Die Betriebsanleitung muss Angaben zu Luftschallemissionen und Vibrationen enthalten.",
+     "norm_ids": [
+      "EU-MaschVO-AnhangIII"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -5089,7 +5206,7 @@
    "member_count": 22,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -5552,5 +5669,13 @@
    ],
    "note": "Common-Criteria-/TOE-/SFR-Evaluierung, BCM, Banking, Smart-Meter-Gateway, DNS und allgemeine ISMS-/OT-Cybersecurity-Themen ohne direkten MaschVO-Bezug; nur teilweise IN-Scope-Anteile bereits in access_control_safety_functions abgebildet"
   }
- ]
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/cra_procedures.json

@@ -11,7 +11,12 @@
    "name": "SBOM-Erstellungsprozess",
    "description": "Erzeugen einer vollstaendigen, maschinenlesbaren Software Bill of Materials fuer ein Produkt mit digitalen Elementen.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["sbom_creation", "sbom_dependency_coverage", "sbom_format_standard", "sbom_tooling_automation"],
+   "fulfills_obligations": [
+    "sbom_creation",
+    "sbom_dependency_coverage",
+    "sbom_format_standard",
+    "sbom_tooling_automation"
+   ],
    "steps": [
     "Komponenten und (direkte + transitive) Abhaengigkeiten inventarisieren",
     "SBOM automatisiert in der Build-/Toolchain generieren",
@@ -24,15 +29,22 @@
     "Format ist maschinenlesbar und standardkonform (CycloneDX/SPDX)",
     "direkte und transitive Abhaengigkeiten enthalten"
    ],
-   "evidence": ["sbom.cyclonedx.json", "Format-Validierungs-Log", "Build-/Toolchain-Konfiguration"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "sbom.cyclonedx.json",
+    "Format-Validierungs-Log",
+    "Build-/Toolchain-Konfiguration"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "sbom_update_process",
    "name": "SBOM-Aktualisierungsprozess",
    "description": "Halten der SBOM aktuell ueber den Produktlebenszyklus bei Komponenten-, Versions- und Patch-Aenderungen.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["sbom_maintenance_update"],
+   "fulfills_obligations": [
+    "sbom_maintenance_update"
+   ],
    "steps": [
     "Komponentenaenderung erkennen (Dependency-/Patch-/Versionsaenderung)",
     "SBOM neu generieren",
@@ -45,15 +57,24 @@
     "SBOM-Version passt zum Release",
     "Supplier-Komponenten enthalten"
    ],
-   "evidence": ["sbom.json", "CI-Log", "Release-Artefakt", "Supplier-SBOM"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "sbom.json",
+    "CI-Log",
+    "Release-Artefakt",
+    "Supplier-SBOM"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "sbom_supplier_integration_process",
    "name": "Lieferanten-SBOM-Integration",
    "description": "Beschaffen und Einarbeiten von Lieferanten-/Drittkomponenten-SBOMs in die Produkt-SBOM.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["sbom_supply_chain_contracts", "sbom_dependency_coverage"],
+   "fulfills_obligations": [
+    "sbom_supply_chain_contracts",
+    "sbom_dependency_coverage"
+   ],
    "steps": [
     "SBOM-Anforderung in Lieferantenvertraege aufnehmen",
     "Lieferanten-SBOMs einsammeln",
@@ -65,15 +86,24 @@
     "Lieferanten-SBOMs eingegangen",
     "Drittkomponenten in der SBOM gelistet"
    ],
-   "evidence": ["Lieferantenvertrag-Klausel", "eingegangene Supplier-SBOMs", "gemergte SBOM"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Lieferantenvertrag-Klausel",
+    "eingegangene Supplier-SBOMs",
+    "gemergte SBOM"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "sbom_provision_process",
    "name": "SBOM-Bereitstellungsprozess",
    "description": "Zugaenglichmachen der SBOM fuer berechtigte Parteien (Nutzer, Behoerde) unter Wahrung der Vertraulichkeit.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["sbom_access_provision", "sbom_authority_provision", "sbom_confidentiality"],
+   "fulfills_obligations": [
+    "sbom_access_provision",
+    "sbom_authority_provision",
+    "sbom_confidentiality"
+   ],
    "steps": [
     "Zugangskanal definieren (Portal/API/dokumentierter Pfad)",
     "Nutzer ueber den Zugangsweg informieren",
@@ -85,15 +115,23 @@
     "Zugriffskontrolle/Vertraulichkeit umgesetzt",
     "Behoerden-Bereitstellungsprozess definiert"
    ],
-   "evidence": ["Zugangskanal-Dokumentation", "Behoerden-Anfrage-Log", "Zugriffskontroll-Konfiguration"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Zugangskanal-Dokumentation",
+    "Behoerden-Anfrage-Log",
+    "Zugriffskontroll-Konfiguration"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "sbom_conformity_documentation_process",
    "name": "SBOM in technischer Dokumentation/Konformitaet",
    "description": "Aufnehmen der SBOM in die technische Dokumentation und Verifizieren der Vollstaendigkeit fuer die Konformitaetsbewertung.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["sbom_technical_documentation", "sbom_completeness_verification"],
+   "fulfills_obligations": [
+    "sbom_technical_documentation",
+    "sbom_completeness_verification"
+   ],
    "steps": [
     "SBOM in die technische Dokumentation aufnehmen",
     "Vollstaendigkeit gegen die real eingesetzte Softwarekomposition pruefen",
@@ -104,16 +142,22 @@
     "Vollstaendigkeit verifiziert",
     "Konformitaetsnachweis vorhanden"
    ],
-   "evidence": ["technische Dokumentation", "Vollstaendigkeits-Pruefbericht", "Konformitaetsnachweis"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "technische Dokumentation",
+    "Vollstaendigkeits-Pruefbericht",
+    "Konformitaetsnachweis"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
-
   {
    "procedure_id": "vuln_handling_process_setup",
    "name": "Schwachstellenbehandlungsprozess einrichten",
    "description": "Dokumentierten Prozess und Meldekanal (CVD) fuer die Schwachstellenbehandlung etablieren.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["vuln_handling_process"],
+   "fulfills_obligations": [
+    "vuln_handling_process"
+   ],
    "steps": [
     "dokumentierten Schwachstellenbehandlungsprozess definieren",
     "Coordinated-Vulnerability-Disclosure-Richtlinie und Meldekanal veroeffentlichen",
@@ -124,15 +168,22 @@
     "Meldekanal/Kontaktstelle auffindbar (z.B. security.txt)",
     "Triage-Verfahren vorhanden"
    ],
-   "evidence": ["Prozessdokument", "security.txt / Kontaktstelle", "Triage-Log"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Prozessdokument",
+    "security.txt / Kontaktstelle",
+    "Triage-Log"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "vuln_identification_process",
    "name": "Schwachstellen-Identifikation",
    "description": "Bekannte Schwachstellen in eingesetzten Komponenten erkennen und inventarisieren.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["vuln_identification_inventory"],
+   "fulfills_obligations": [
+    "vuln_identification_inventory"
+   ],
    "steps": [
     "Advisories/CVE-Feeds beobachten",
     "gegen die SBOM-Komponenten abgleichen",
@@ -143,15 +194,21 @@
     "SBOM-zu-CVE-Abgleich durchgefuehrt",
     "Schwachstellen-Inventar gepflegt"
    ],
-   "evidence": ["CVE-Abgleich-Report", "Schwachstellen-Register"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "CVE-Abgleich-Report",
+    "Schwachstellen-Register"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "vuln_assessment_process",
    "name": "Schwachstellen-Bewertung/Priorisierung",
    "description": "Identifizierte Schwachstellen nach Schweregrad, Ausnutzbarkeit und Exposition bewerten und priorisieren.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["vuln_assessment_prioritization"],
+   "fulfills_obligations": [
+    "vuln_assessment_prioritization"
+   ],
    "steps": [
     "Schweregrad bewerten (z.B. CVSS)",
     "Ausnutzbarkeit/Exposition einschaetzen",
@@ -161,15 +218,21 @@
     "Schweregrad standardisiert bewertet",
     "risikobasierte Priorisierung vorhanden"
    ],
-   "evidence": ["Bewertungsdatensatz (CVSS)", "Prioritaetenliste"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Bewertungsdatensatz (CVSS)",
+    "Prioritaetenliste"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "vuln_remediation_process",
    "name": "Schwachstellen-Behebung",
    "description": "Bekannte Schwachstellen fristgerecht durch Patches/Gegenmassnahmen beheben und Sicherheitsupdates bereitstellen.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["vuln_remediation_patching"],
+   "fulfills_obligations": [
+    "vuln_remediation_patching"
+   ],
    "steps": [
     "Fix/Gegenmassnahme entwickeln",
     "testen",
@@ -181,15 +244,23 @@
     "Sicherheitsupdate bereitgestellt",
     "Follow-up bis Closure"
    ],
-   "evidence": ["Patch/Release", "Behebungs-Zeitleiste", "Follow-up-Log"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Patch/Release",
+    "Behebungs-Zeitleiste",
+    "Follow-up-Log"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "vuln_disclosure_process",
    "name": "Offenlegung + Nutzerinformation",
    "description": "Koordinierte Offenlegung behobener Schwachstellen und Information der Nutzer ueber Schutzmassnahmen.",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["coordinated_vulnerability_disclosure", "vuln_info_dissemination_users"],
+   "fulfills_obligations": [
+    "coordinated_vulnerability_disclosure",
+    "vuln_info_dissemination_users"
+   ],
    "steps": [
     "Offenlegungszeitpunkt koordinieren",
     "Security Advisory / CVE-Eintrag veroeffentlichen",
@@ -199,15 +270,22 @@
     "Advisory veroeffentlicht",
     "Nutzer informiert"
    ],
-   "evidence": ["Security Advisory", "CVE-Eintrag", "Nutzer-Benachrichtigung"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "Security Advisory",
+    "CVE-Eintrag",
+    "Nutzer-Benachrichtigung"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
   },
   {
    "procedure_id": "vuln_authority_reporting_process",
    "name": "Behoerdenmeldung aktiv ausgenutzter Schwachstellen",
    "description": "Aktiv ausgenutzte Schwachstellen fristgerecht an CSIRT/ENISA melden (CRA Art. 14-Kaskade).",
    "source_role": "procedural_requirement",
-   "fulfills_obligations": ["exploited_vuln_reporting_authorities"],
+   "fulfills_obligations": [
+    "exploited_vuln_reporting_authorities"
+   ],
    "applicability_note": "bedingt: nur bei aktiv ausgenutzter Schwachstelle",
    "steps": [
     "aktive Ausnutzung erkennen",
@@ -220,8 +298,20 @@
     "72h-Meldung erfolgt",
     "14d-Abschlussbericht erfolgt"
    ],
-   "evidence": ["CSIRT/ENISA-Meldungsbelege", "Zeitstempel der Kaskade"],
-   "citation_spans": [], "citation_status": "pending_span_anchor"
+   "evidence": [
+    "CSIRT/ENISA-Meldungsbelege",
+    "Zeitstempel der Kaskade"
+   ],
+   "citation_spans": [],
+   "citation_status": "pending_span_anchor"
+  }
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
  }
- ]
 }

obligations/cra_remote_access.json

@@ -46,7 +46,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(2)(d)",
-     "citation": "Schutz vor unbefugtem Zugriff durch geeignete Kontrollmechanismen (Authentifizierung, Identitaets- und Zugriffsmanagement)"
+     "citation": "Schutz vor unbefugtem Zugriff durch geeignete Kontrollmechanismen (Authentifizierung, Identitaets- und Zugriffsmanagement)",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -342,7 +346,7 @@
    "member_count": 277,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.92,
@@ -370,7 +374,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(2)(b)(c)",
-     "citation": "Schutz der Vertraulichkeit und Integritaet von Daten und Befehlen"
+     "citation": "Schutz der Vertraulichkeit und Integritaet von Daten und Befehlen",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -656,7 +664,7 @@
    "member_count": 274,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -917,7 +925,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(2)(g)",
-     "citation": "Aufzeichnung und Ueberwachung relevanter interner Aktivitaeten (Logging)"
+     "citation": "Aufzeichnung und Ueberwachung relevanter interner Aktivitaeten (Logging)",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -959,7 +971,7 @@
    "member_count": 22,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1145,7 +1157,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(2)(a)",
-     "citation": "Bereitstellung ohne bekannte ausnutzbare Schwachstellen / minimierte Angriffsflaeche"
+     "citation": "Bereitstellung ohne bekannte ausnutzbare Schwachstellen / minimierte Angriffsflaeche",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1178,7 +1194,7 @@
    "member_count": 19,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.83,
@@ -1210,7 +1226,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (2)(1)",
-     "citation": "Behandlung und Behebung von Schwachstellen, Sicherheitsupdates"
+     "citation": "Behandlung und Behebung von Schwachstellen, Sicherheitsupdates",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1247,7 +1267,7 @@
    "member_count": 17,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.82,
@@ -1662,5 +1682,13 @@
    ],
    "note": "Physische Maschinen-Fernsteuerung/Ergonomie/Gefahrenzonen-Sicherheit (MaschinenVO 2023/1230), keine Cybersecurity-Fernwartung"
   }
- ]
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/cra_updates.json

@@ -52,12 +52,20 @@
     {
      "source": "CRA",
      "anchor": "Annex I (2)(c)",
-     "citation": "Schwachstellen durch Sicherheitsupdates ohne Verzug behandeln, einschliesslich automatischer Updates und Benachrichtigung."
+     "citation": "Schwachstellen durch Sicherheitsupdates ohne Verzug behandeln, einschliesslich automatischer Updates und Benachrichtigung.",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     },
     {
      "source": "CRA",
      "anchor": "Art. 13",
-     "citation": "Pflicht zur Bereitstellung von Sicherheitsupdates waehrend des Support-Zeitraums."
+     "citation": "Pflicht zur Bereitstellung von Sicherheitsupdates waehrend des Support-Zeitraums.",
+     "norm_ids": [
+      "EU-CRA-Art13"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [
@@ -661,7 +669,7 @@
    "member_count": 578,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.95,
@@ -689,7 +697,11 @@
     {
      "source": "CRA",
      "anchor": "Art. 13(8)",
-     "citation": "Bestimmung des Support-Zeitraums entsprechend der erwarteten Nutzungsdauer."
+     "citation": "Bestimmung des Support-Zeitraums entsprechend der erwarteten Nutzungsdauer.",
+     "norm_ids": [
+      "EU-CRA-Art13"
+     ],
+     "norm_id_status": "article_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1275,7 +1287,7 @@
    "member_count": 574,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -1303,7 +1315,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(3)(f)",
-     "citation": "Schutz der Integritaet von Daten, Befehlen und Konfigurationen vor Manipulation."
+     "citation": "Schutz der Integritaet von Daten, Befehlen und Konfigurationen vor Manipulation.",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1382,7 +1398,7 @@
    "member_count": 58,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1415,7 +1431,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(3)(d)",
-     "citation": "Schutz vor unbefugtem Zugriff durch geeignete Kontrollmechanismen."
+     "citation": "Schutz vor unbefugtem Zugriff durch geeignete Kontrollmechanismen.",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [
@@ -1476,7 +1496,7 @@
    "member_count": 42,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.85,
@@ -1642,7 +1662,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (2)(c)",
-     "citation": "Sicherheitsupdates werden, soweit moeglich, automatisch installiert mit Opt-out-Moeglichkeit des Nutzers."
+     "citation": "Sicherheitsupdates werden, soweit moeglich, automatisch installiert mit Opt-out-Moeglichkeit des Nutzers.",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1661,7 +1685,7 @@
    "member_count": 6,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.9,
@@ -1690,7 +1714,11 @@
     {
      "source": "CRA",
      "anchor": "Annex I (1)(2)",
-     "citation": "Cybersicherheits-Risikobeurteilung als Grundlage fuer Schwachstellenbehandlung."
+     "citation": "Cybersicherheits-Risikobeurteilung als Grundlage fuer Schwachstellenbehandlung.",
+     "norm_ids": [
+      "EU-CRA-AnhangI"
+     ],
+     "norm_id_status": "annex_confirmed"
     }
    ],
    "guidance_basis": [],
@@ -1704,7 +1732,7 @@
    "member_count": 2,
    "relationships": [],
    "citation_anchor_ids": [],
-   "citation_status": "pending_span_anchor",
+   "citation_status": "norm_id_linked",
    "review_status": "draft",
    "provenance": {
     "discovery_confidence": 0.8,
@@ -1816,5 +1844,13 @@
    ],
    "note": "M4 (digitale Veraenderungen allgemein) und M7 (TLS-Proxy-Kanalverwaltung) betreffen Konfigurations-/Netzwerkmanagement, nicht die Update-/Patch-Pflicht im engeren Sinne."
   }
- ]
+ ],
+ "norm_id_contract": {
+  "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+  "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+  "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+  "article_status": "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed",
+  "source": "Board Compliance/KB-v2 2026-07-01",
+  "kb_v2_verification": "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); 3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+ }
 }

obligations/norm_id_manifest.json

@@ -0,0 +1,43 @@
+{
+ "contract": "norm_id logical citation join (Domäne 2 legal_basis -> KB-v2 units)",
+ "convention": "EU-<ACT>-Anhang<ROM> (annex) / EU-<ACT>-Art<N> (article) — beide in KB-v2 confirmed; EU-<ACT>-Kapitel<ROM> NICHT gemintet",
+ "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+ "granularity": "annex-grob",
+ "source": "KB-v2 bp_compliance_kb_2026_1_build",
+ "kb_v2_verification": {
+  "date": "2026-07-01",
+  "by": "Compliance/KB (RAG-Ingestion)-Session",
+  "result": "16/19 verify_pending bestätigt (alle Artikel existieren); 3 Kapitel-IDs fehlen (Compiler mintet Artikel+Annex, keine Kapitel)",
+  "resolution": "Kapitel re-anchoren auf Konstituenten-Artikel (KB-v2 hat alle) ODER Kapitel-Units als KB-Enhancement; hier als chapter_no_kb_unit markiert, NICHT geraten"
+ },
+ "annex_confirmed": {
+  "EU-CRA-AnhangI": 34,
+  "EU-CRA-AnhangVII": 2,
+  "EU-MaschVO-AnhangI": 2,
+  "EU-MaschVO-AnhangIII": 14,
+  "EU-MaschVO-AnhangIV": 2
+ },
+ "article_confirmed": {
+  "EU-CRA-Art13": 3,
+  "EU-CRA-Art14": 1,
+  "EU-CRA-Art16": 1,
+  "EU-CRA-Art3": 1,
+  "EU-CRA-Art31": 3,
+  "EU-MaschVO-Art1": 1,
+  "EU-MaschVO-Art10": 1,
+  "EU-MaschVO-Art11": 1,
+  "EU-MaschVO-Art18": 1,
+  "EU-MaschVO-Art21": 1,
+  "EU-MaschVO-Art22": 1,
+  "EU-MaschVO-Art25": 1,
+  "EU-MaschVO-Art5": 1,
+  "EU-MaschVO-Art50": 1,
+  "EU-MaschVO-Art53": 1,
+  "EU-MaschVO-Art54": 1
+ },
+ "chapter_no_kb_unit": {
+  "EU-MaschVO-KapitelIV": 1,
+  "EU-MaschVO-KapitelV": 1,
+  "EU-MaschVO-KapitelVI": 1
+ }
+}

obligations/scope_audit_findings.json

@@ -0,0 +1,38 @@
+{
+ "audit": "obligation scope audit (Adressat: Hersteller vs Behörde/notified_body)",
+ "principle": "Adressat der Norm != Handlungspflicht des Herstellers; scope-Achse in_scope/out_of_scope/derived_obligation",
+ "false_positive_guard": "Melde-AN-Behörde-Pflichten (applicability=domain:products…) bleiben IN-SCOPE",
+ "obligations_scanned": 126,
+ "classified": [
+  {
+   "file": "cra_machinery.json",
+   "id": "notified_body_requirements",
+   "name": "Anforderungen an notifizierte Stellen",
+   "tier": "LEGAL_MINIMUM",
+   "applicability": "domain:notified_body",
+   "scope": "derived_obligation",
+   "scope_reason": "Norm adressiert primär die notifizierte Stelle (Unabhängigkeit/Kompetenz/Unparteilichkeit), erzeugt aber mittelbare Hersteller-Pflichten: notifizierte Stelle einbeziehen, erforderliche Unterlagen bereitstellen, Konformitätsbewertung korrekt durchführen.",
+   "scope_split_note": "Kandidat für spätere Aufspaltung: 'Normadressat' (Anforderungen AN die notifizierte Stelle = institutional/out_of_scope) ↔ 'abgeleitete Herstellerpflicht' (NB einbeziehen + Unterlagen + Konformitätsbewertung = in_scope). NICHT vorzeitig festziehen."
+  },
+  {
+   "file": "cra_machinery.json",
+   "id": "market_surveillance_safeguard",
+   "name": "Marktüberwachung, nationale Schutzmaßnahmen und Korrekturmaßnahmen",
+   "tier": "LEGAL_MINIMUM",
+   "applicability": "domain:authority",
+   "scope": "out_of_scope",
+   "scope_reason": "Adressat = Marktüberwachungsbehörden/Kommission (Schutzmaßnahmen, Schutzklauselverfahren); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA."
+  },
+  {
+   "file": "cra_machinery.json",
+   "id": "sanctions",
+   "name": "Sanktionen für Verstöße gegen die Maschinenverordnung",
+   "tier": "LEGAL_MINIMUM",
+   "applicability": "domain:authority",
+   "scope": "out_of_scope",
+   "scope_reason": "Adressat = Mitgliedstaaten (legen Sanktionen fest); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA (CRA-Vuln-Cut)."
+  }
+ ],
+ "unclassified_candidates": [],
+ "decision_owner": "User/Registry-Owner — Audit FLAGGT nur; für jeden künftigen Cut mitlaufen lassen"
+}

scripts/obligation_discovery/apply_norm_id_join.py

@@ -0,0 +1,75 @@
+"""Zitierfähigkeits-Join (Wake-up #2): logischer norm_id-Join auf legal_basis.
+
+KB-v2-Konvention (Board 2026-07-01, Compliance/KB-v2): `EU-<ACT>-Anhang<ROM>` (Annex-Ebene, grob) ·
+`EU-<ACT>-Art<N>` (Artikel, in KB-v2 noch zu verifizieren) · Kapitel = TBD-Konvention.
+Namensvariante: `EU-MaschVO-*` (NICHT MaschinenVO). Kein char-Span nötig — logischer Join auf norm_id.
+Fügt `norm_ids` (Liste) je legal_basis + `norm_id_status` hinzu; setzt obligation.citation_status
+auf `norm_id_linked` (annex-grob). KEINE neue Klasse (Attribut). Freeze-safe.
+"""
+from __future__ import annotations
+
+import glob
+import json
+import re
+
+ACT = {"CRA": "CRA", "MaschVO": "MaschVO", "MaschinenVO": "MaschVO"}
+FILES = sorted(glob.glob("obligations/cra*.json"))
+
+
+def derive(source: str, anchor: str) -> tuple[list[str], str]:
+    act = ACT.get(source, source)
+    ids: list[str] = []
+    for rom in re.findall(r"An(?:hang|nex)\s+([IVX]+)", anchor, re.I):
+        ids.append(f"EU-{act}-Anhang{rom.upper()}")
+    articles = re.findall(r"\bArt(?:icle|\.)?\s*(\d+)", anchor)
+    chapters = re.findall(r"Kapitel\s+([IVX/]+)", anchor, re.I)
+    verify: list[str] = []
+    for n in articles:
+        verify.append(f"EU-{act}-Art{n}")
+    for grp in chapters:
+        for rom in grp.split("/"):
+            rom = rom.strip()
+            if rom:
+                verify.append(f"EU-{act}-Kapitel{rom.upper()}")
+    # dedup, Annexe zuerst (confirmed), dann verify
+    seen: set[str] = set()
+    ordered = [x for x in ids + verify if not (x in seen or seen.add(x))]
+    status = "annex_confirmed" if ids else ("verify_pending" if verify else "unparsed")
+    return ordered, status
+
+
+def main() -> None:
+    total_lb = linked = unparsed = 0
+    obl_linked = 0
+    for f in FILES:
+        d = json.load(open(f, encoding="utf-8"))
+        d.setdefault("norm_id_contract", {
+            "convention": "EU-<ACT>-Anhang<ROM> (Annex-Ebene) / EU-<ACT>-Art<N> (verify) — KB-v2 bp_compliance_kb_2026_1_build",
+            "act_naming": "EU-MaschVO-* (NICHT MaschinenVO)",
+            "granularity": "annex-grob — 'Annex I Part II (1)' -> EU-CRA-AnhangI; Part/Punkt = KB-Enhancement TBD",
+            "article_status": "EU-<ACT>-Art<N> in KB-v2 noch zu verifizieren; Annex-IDs confirmed",
+            "source": "Board Compliance/KB-v2 2026-07-01",
+        })
+        for o in d.get("obligations", []):
+            got = False
+            for b in o.get("legal_basis", []):
+                total_lb += 1
+                nids, st = derive(b.get("source", ""), b.get("anchor", ""))
+                b["norm_ids"] = nids
+                b["norm_id_status"] = st
+                if nids:
+                    linked += 1
+                    got = True
+                if st == "unparsed":
+                    unparsed += 1
+                    print(f"  UNPARSED: {b.get('source')} \"{b.get('anchor')}\"")
+            if got:
+                o["citation_status"] = "norm_id_linked"
+                obl_linked += 1
+        json.dump(d, open(f, "w", encoding="utf-8"), ensure_ascii=False, indent=1)
+    print(f"legal_basis gesamt {total_lb} | mit norm_ids {linked} | unparsed {unparsed}")
+    print(f"Obligations citation_status -> norm_id_linked: {obl_linked}")
+
+
+if __name__ == "__main__":
+    main()

scripts/obligation_discovery/apply_norm_id_verification.py

@@ -0,0 +1,70 @@
+"""KB-v2-Verifikation der norm_ids einarbeiten (2026-07-01, Feedback Compliance/KB-Session).
+
+KB-v2 hat die 19 verify_pending IDs geprüft: 16/19 ✓ (alle CRA+MaschVO-Artikel existieren),
+3 fehlen = Kapitel-Ebene (`EU-MaschVO-KapitelIV/V/VI`) — der KB-Compiler mintet Artikel+Annex,
+KEINE Kapitel. Konsequenz: Artikel-norm_ids `verify_pending→article_confirmed`; Kapitel-norm_ids
+`chapter_no_kb_unit` (danglender Join-Key) + Re-Anchor-Hinweis (KB-v2 hat die Konstituenten-Artikel;
+Re-Anchor = Enhancement, NICHT geraten). Deterministisch aus dem norm_id-Inhalt neu abgeleitet.
+"""
+from __future__ import annotations
+
+import glob
+import json
+
+REANCHOR_NOTE = (
+    "Kapitel-Ebene nicht als KB-v2-Unit gemintet (Compiler = Artikel+Annex). "
+    "Re-Anchor auf Konstituenten-Artikel = Enhancement (KB-v2 hat die Artikel); NICHT geraten."
+)
+
+
+def status_for(norm_ids: list[str]) -> str:
+    has_annex = any("-Anhang" in n for n in norm_ids)
+    has_article = any("-Art" in n and "-Anhang" not in n for n in norm_ids)
+    has_chapter = any("-Kapitel" in n for n in norm_ids)
+    if has_annex:
+        return "annex_confirmed"
+    if has_article:
+        return "article_confirmed"  # KB-v2 verified 16/16
+    if has_chapter:
+        return "chapter_no_kb_unit"
+    return "unparsed"
+
+
+def main() -> None:
+    counts = {"annex_confirmed": 0, "article_confirmed": 0, "chapter_no_kb_unit": 0}
+    obl_linked = obl_chapter = 0
+    for f in sorted(glob.glob("obligations/cra*.json")):
+        d = json.load(open(f, encoding="utf-8"))
+        for o in d.get("obligations", []):
+            joinable = chapter_only = False
+            for b in o.get("legal_basis", []):
+                nids = b.get("norm_ids", [])
+                st = status_for(nids)
+                b["norm_id_status"] = st
+                counts[st] = counts.get(st, 0) + 1
+                if st == "chapter_no_kb_unit":
+                    b["norm_id_note"] = REANCHOR_NOTE
+                    chapter_only = True
+                elif st in ("annex_confirmed", "article_confirmed"):
+                    joinable = True
+            if joinable:
+                o["citation_status"] = "norm_id_linked"
+                obl_linked += 1
+            elif chapter_only:
+                o["citation_status"] = "chapter_reanchor_pending"
+                obl_chapter += 1
+        # Contract-Block um den Verifikationsstand ergänzen
+        c = d.get("norm_id_contract")
+        if isinstance(c, dict):
+            c["kb_v2_verification"] = (
+                "2026-07-01: 16/19 verify_pending IDs in KB-v2 bestätigt (alle Artikel); "
+                "3 Kapitel-IDs = chapter_no_kb_unit (Compiler mintet keine Kapitel)."
+            )
+            c["article_status"] = "EU-<ACT>-Art<N> in KB-v2 BESTÄTIGT (16/16); Annex-IDs confirmed"
+        json.dump(d, open(f, "w", encoding="utf-8"), ensure_ascii=False, indent=1)
+    print("legal_basis status:", counts)
+    print(f"citation_status: norm_id_linked {obl_linked} | chapter_reanchor_pending {obl_chapter}")
+
+
+if __name__ == "__main__":
+    main()

scripts/obligation_discovery/apply_scope_classification.py

@@ -0,0 +1,53 @@
+"""Scope-Klassifikation anwenden (User-Entscheidung 2026-07-01, Option 2 + derived_obligation).
+
+Neue `scope`-Attribut-Achse (KEINE neue Objektklasse — Enum-Wert, freeze-safe):
+  in_scope (default/implizit) · out_of_scope · derived_obligation
+
+Prinzip (User): Adressat der Norm ⊥ Handlungspflicht des Herstellers. Reine Staats-/
+Durchsetzungs-/Institutions-Bestimmungen = out_of_scope. Norm, die primär eine andere Rolle
+adressiert ABER mittelbar eine Hersteller-Handlungspflicht erzeugt = derived_obligation
+(bleibt im Hersteller-Set, wird NICHT verworfen — 'im Zweifel nicht zu früh Wissen verwerfen').
+"""
+from __future__ import annotations
+
+import glob
+import json
+
+SCOPE = {
+    "sanctions": {
+        "scope": "out_of_scope",
+        "scope_reason": "Adressat = Mitgliedstaaten (legen Sanktionen fest); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA (CRA-Vuln-Cut).",
+    },
+    "market_surveillance_safeguard": {
+        "scope": "out_of_scope",
+        "scope_reason": "Adressat = Marktüberwachungsbehörden/Kommission (Schutzmaßnahmen, Schutzklauselverfahren); keine Hersteller-Handlungspflicht. Präzedenz CSIRT/ENISA.",
+    },
+    "notified_body_requirements": {
+        "scope": "derived_obligation",
+        "scope_reason": "Norm adressiert primär die notifizierte Stelle (Unabhängigkeit/Kompetenz/Unparteilichkeit), erzeugt aber mittelbare Hersteller-Pflichten: notifizierte Stelle einbeziehen, erforderliche Unterlagen bereitstellen, Konformitätsbewertung korrekt durchführen.",
+        "scope_split_candidate": True,
+        "scope_split_note": "Kandidat für spätere Aufspaltung: 'Normadressat' (Anforderungen AN die notifizierte Stelle = institutional/out_of_scope) ↔ 'abgeleitete Herstellerpflicht' (NB einbeziehen + Unterlagen + Konformitätsbewertung = in_scope). NICHT vorzeitig festziehen.",
+    },
+}
+
+
+def main() -> None:
+    applied = []
+    for f in sorted(glob.glob("obligations/cra*.json")):
+        d = json.load(open(f, encoding="utf-8"))
+        changed = False
+        for o in d.get("obligations", []):
+            spec = SCOPE.get(o.get("id"))
+            if spec:
+                o.update(spec)
+                applied.append((o["id"], spec["scope"]))
+                changed = True
+        if changed:
+            json.dump(d, open(f, "w", encoding="utf-8"), ensure_ascii=False, indent=1)
+    for oid, sc in applied:
+        print(f"  {oid:32} scope={sc}")
+    print(f"angewendet: {len(applied)} (erwartet 3)")
+
+
+if __name__ == "__main__":
+    main()

scripts/obligation_discovery/export_join_keys.py

@@ -23,12 +23,15 @@ def main() -> None:
     for path in a.registries:
         reg = json.load(open(path, encoding="utf-8"))
         for o in reg.get("obligations", []):
+            if o.get("scope") == "out_of_scope":
+                continue  # Institutions-/Enforcement-Recht (Adressat != Hersteller) -> nicht im Join-Vertrag
             citation_units = [b.get("anchor", "") for b in o.get("legal_basis", []) if b.get("anchor")]
             keys.append({
                 "obligation_id": o["id"],
                 "regulation": reg.get("regulation", ""),
                 "family": o.get("family", ""),
                 "tier": o.get("tier", ""),
+                "scope": o.get("scope", "in_scope"),
                 "citation_units": citation_units,
                 "source_role": o.get("source_role", ""),
             })

scripts/obligation_discovery/scope_audit.py

@@ -0,0 +1,74 @@
+"""Scope-Audit: Adressaten-Prüfung der Obligation-Registry (Review-Stage-Werkzeug).
+
+Prinzip (User 2026-07-01): **Adressat der Norm ⊥ Handlungspflicht des Herstellers.** Die Registry
+modelliert Hersteller-Pflichten. `scope`-Achse (Attribut-Enum, KEINE neue Objektklasse — freeze-safe):
+  - in_scope           : Norm adressiert direkt den Hersteller (default).
+  - out_of_scope       : reines Staats-/Durchsetzungs-/Institutions-Recht (Adressat != Hersteller,
+                         KEINE mittelbare Herstellerpflicht) — z.B. Sanktionen, Marktüberwachung.
+                         Präzedenz CSIRT/ENISA (CRA-Vuln-Cut). Aus join_keys gefiltert.
+  - derived_obligation : Norm adressiert primär eine andere Rolle, erzeugt aber MITTELBAR eine
+                         Hersteller-Handlungspflicht — bleibt im Set (Wissen nicht verwerfen),
+                         ggf. `scope_split_candidate` (Aufspaltung Normadressat ↔ abgeleitete Pflicht).
+
+False-Positive-Guard: Melde-AN-Behörde-Pflichten (applicability=domain:products…) sind IN-SCOPE
+(Adressat = Hersteller, Behörde nur Empfänger) — der Audit key't auf `applicability`, nicht auf
+Behörden-Nennung im Namen. Deterministisch, kein LLM. Audit FLAGGT; Reklassifizierung = User/Owner.
+"""
+from __future__ import annotations
+
+import glob
+import json
+
+NON_MANUFACTURER_DOMAINS = {
+    "domain:authority",
+    "domain:notified_body",
+    "domain:market_surveillance",
+    "domain:member_state",
+    "domain:commission",
+}
+
+
+def main() -> None:
+    classified = []          # bereits per scope entschieden
+    unclassified = []        # nicht-Hersteller-Adressat OHNE scope -> Review-Kandidat
+    total = 0
+    for f in sorted(glob.glob("obligations/cra*.json")):
+        d = json.load(open(f, encoding="utf-8"))
+        for o in d.get("obligations", []):
+            total += 1
+            appl = (o.get("applicability") or "").strip()
+            scope = o.get("scope")
+            rec = {
+                "file": f.split("/")[-1], "id": o.get("id") or o.get("obligation_id"),
+                "name": o.get("name"), "tier": o.get("tier"), "applicability": appl,
+                "scope": scope, "scope_reason": o.get("scope_reason"),
+            }
+            if scope in ("out_of_scope", "derived_obligation"):
+                if o.get("scope_split_candidate"):
+                    rec["scope_split_note"] = o.get("scope_split_note")
+                classified.append(rec)
+            elif appl in NON_MANUFACTURER_DOMAINS:
+                rec["reason"] = "Adressat Behörde/notifizierte Stelle/Mitgliedstaat, nicht klassifiziert"
+                rec["recommendation"] = "out_of_scope (reine Durchsetzung) ODER derived_obligation (mittelbare Herstellerpflicht)"
+                unclassified.append(rec)
+    out = {
+        "audit": "obligation scope audit (Adressat: Hersteller vs Behörde/notified_body)",
+        "principle": "Adressat der Norm != Handlungspflicht des Herstellers; scope-Achse in_scope/out_of_scope/derived_obligation",
+        "false_positive_guard": "Melde-AN-Behörde-Pflichten (applicability=domain:products…) bleiben IN-SCOPE",
+        "obligations_scanned": total,
+        "classified": classified,
+        "unclassified_candidates": unclassified,
+        "decision_owner": "User/Registry-Owner — Audit FLAGGT nur; für jeden künftigen Cut mitlaufen lassen",
+    }
+    json.dump(out, open("obligations/scope_audit_findings.json", "w", encoding="utf-8"),
+              ensure_ascii=False, indent=1)
+    print(f"gescannt {total} | klassifiziert {len(classified)} | unklassifizierte Kandidaten {len(unclassified)}")
+    for r in classified:
+        extra = " [SPLIT-KANDIDAT]" if r.get("scope_split_note") else ""
+        print(f"  [{r['scope']}] {r['id']}  appl={r['applicability']}{extra}")
+    for r in unclassified:
+        print(f"  [UNKLASSIFIZIERT] {r['id']}  appl={r['applicability']}")
+
+
+if __name__ == "__main__":
+    main()

scripts/obligation_discovery/validate_registry.py

@@ -11,6 +11,7 @@ import json
 import sys
 
 from _core import validate_registry
+from scope_audit import NON_MANUFACTURER_DOMAINS
 
 
 def main() -> None:
@@ -28,6 +29,15 @@ def main() -> None:
     print(f"  out_of_scope:              {v['out_of_scope']}")
     print(f"  semantische Kanten:        {v['semantic_edges']}")
     print(f"  PASSED: {v['passed']}")
+    # Scope-Audit (Review-Step, FLAG-ONLY — mutiert nie, blockt nie, ändert Exit-Code nicht):
+    unclassified = [o.get("id") or o.get("obligation_id") for o in reg.get("obligations", [])
+                    if (o.get("applicability") or "").strip() in NON_MANUFACTURER_DOMAINS and not o.get("scope")]
+    if unclassified:
+        print(f"  ⚠ SCOPE-AUDIT: {len(unclassified)} authority-/institution-adressierte Obligation(s) OHNE scope:")
+        print(f"      {', '.join(str(x) for x in unclassified)}")
+        print("      → Review-Pflicht: scope=out_of_scope|derived_obligation (apply_scope_classification.py, NUR mit Review-Go). Auto-Reclassification VERBOTEN.")
+    else:
+        print("  Scope-Audit:               keine unklassifizierten authority-/institution-Obligations")
     sys.exit(0 if v["passed"] else 1)