feat(ai-sdk): pin accepted proposer decisions into the GT gate (P3)

When a human accepts a proposer proposal, an AcceptedPin records a machine-scoped
invariant — a pattern MUST fire (coverage/vocab→tag) or must NOT fire
(dedup/framing) — that a test re-checks on every run. This makes the library's
growth COMPOUND into the gate instead of eroding it: a change that re-introduces
a dropped duplicate, un-gates a foreign pattern, or removes a coverage hazard
breaks a pin and fails CI. One boolean covers all four proposal types.

Seeded testdata/accepted_pins_warewashing.json with the accepted P1 supersessions
(HP016/HP018/HP013 must NOT fire; their clean equivalents HP2201/HP144 must fire).
TestWarewashing_AcceptedPins re-checks 5/5 against the live engine output;
GenerateDedupPin turns an accepted dedup verdict into its pin.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

admin-compliance/app/sdk/onboarding-advisor/page.tsx

@@ -0,0 +1,200 @@
+'use client'
+
+// ETO / Onboarding-Advisor — thin operator surface over POST /api/compliance/onboarding/advisor-start.
+// Certifications + target + scanner findings -> Silent Pass -> Advisor. NOT the regulation gap engine
+// (/sdk/gap-analysis is a different flow: product -> applicable regulations). This tests the cert->delta
+// case: "TISAX/ISO27001 -> CRA, what is auto-detected, what stays an open question?". No new backend.
+
+import React, { useEffect, useState } from 'react'
+
+const CERTS = ['ISO27001', 'TISAX', 'ISO9001', 'IEC62443', 'ISO13485', 'ISO14001', 'ASPICE', 'IATF16949']
+
+// label -> {signal_id, source_type} — demonstrates all three signal KINDS (observation / partial / requirement)
+const FINDINGS: Array<{ label: string; signal_id: string; source_type: string; kind: string }> = [
+  { label: 'SBOM im Repo (CycloneDX/SPDX)', signal_id: 'cyclonedx_found', source_type: 'repository', kind: 'observation' },
+  { label: 'security.txt / CVD-Policy veröffentlicht', signal_id: 'security_txt', source_type: 'website', kind: 'observation' },
+  { label: 'Signierte Releases', signal_id: 'signed_releases', source_type: 'repository', kind: 'observation' },
+  { label: 'Produkt-Risikobewertung (Dokument)', signal_id: 'risk_assessment_pdf', source_type: 'document', kind: 'observation' },
+  { label: 'CI-Pipeline vorhanden (nur Indikation)', signal_id: 'github_actions_ci', source_type: 'repository', kind: 'partial' },
+  { label: 'Cloud-/vernetztes Produkt', signal_id: 'cloud_hosted', source_type: 'product', kind: 'observation' },
+  { label: 'Ausschreibung FORDERT SBOM (Requirement)', signal_id: 'requires_sbom', source_type: 'tender', kind: 'requirement' },
+  { label: 'OEM FORDERT PSIRT (Requirement)', signal_id: 'supplier_requires_psirt', source_type: 'oem', kind: 'requirement' },
+]
+
+interface Question { capability_id: string; question_intent: string; why: string; information_value: number; priority: string }
+interface Inferred { certification: string; capabilities: string[]; statement: string }
+interface Rejected { certification?: string; statement: string; reason: string }
+interface Measure { capability_id: string; leverage: number; closes: string[] }
+interface AdvisorResponse {
+  silent_intake_summary: string; headline: string; auto_detected: string[]; indications: string[]
+  inferred_assumptions: Inferred[]; rejected_assumptions: Rejected[]; top_5_questions: Question[]
+  capability_delta: string[]; top_measures: Measure[]; evidence_requests: string[]
+  unsupported_domains: string[]; completeness_summary: string; capability_labels: Record<string, string>
+}
+
+const PROXY = '/api/sdk/v1/compliance/onboarding'
+
+function Chips({ items, tone }: { items: string[]; tone: string }) {
+  if (!items.length) return <span className="text-gray-400 text-sm">—</span>
+  return (
+    <div className="flex flex-wrap gap-2">
+      {items.map(c => <span key={c} className={`px-2.5 py-1 rounded-full text-xs font-medium ${tone}`}>{c}</span>)}
+    </div>
+  )
+}
+
+function Section({ title, hint, children }: { title: string; hint?: string; children: React.ReactNode }) {
+  return (
+    <div className="bg-white rounded-xl border border-gray-200 p-5">
+      <h3 className="font-semibold text-gray-900">{title}</h3>
+      {hint && <p className="text-xs text-gray-500 mt-0.5 mb-2">{hint}</p>}
+      <div className="mt-2">{children}</div>
+    </div>
+  )
+}
+
+export default function OnboardingAdvisorPage() {
+  const [targets, setTargets] = useState<string[]>([])
+  const [company, setCompany] = useState('Beispiel Maschinenbau')
+  const [industry, setIndustry] = useState('machine_builder')
+  const [certs, setCerts] = useState<string[]>(['ISO27001', 'ISO9001'])
+  const [target, setTarget] = useState('CRA')
+  const [findings, setFindings] = useState<string[]>(['cyclonedx_found', 'github_actions_ci', 'requires_sbom'])
+  const [knownEvidence, setKnownEvidence] = useState('CE-Prozess')
+  const [result, setResult] = useState<AdvisorResponse | null>(null)
+  const [loading, setLoading] = useState(false)
+  const [error, setError] = useState('')
+
+  useEffect(() => {
+    fetch(`${PROXY}/targets`).then(r => r.json()).then(d => {
+      if (Array.isArray(d.targets)) { setTargets(d.targets); if (!d.targets.includes('CRA') && d.targets[0]) setTarget(d.targets[0]) }
+    }).catch(() => {})
+  }, [])
+
+  const toggle = (list: string[], set: (v: string[]) => void, v: string) =>
+    set(list.includes(v) ? list.filter(x => x !== v) : [...list, v])
+
+  const lbl = (id: string) => result?.capability_labels?.[id] || id.replace(/_/g, ' ')
+
+  const run = async () => {
+    setLoading(true); setError(''); setResult(null)
+    try {
+      const scanner_findings = FINDINGS.filter(f => findings.includes(f.signal_id))
+        .map(f => ({ signal_id: f.signal_id, source_type: f.source_type }))
+      const res = await fetch(`${PROXY}/advisor-start`, {
+        method: 'POST', headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          company, industry, products: [], markets: ['EU'], certifications: certs,
+          known_evidence: knownEvidence ? knownEvidence.split(',').map(s => s.trim()).filter(Boolean) : [],
+          target, scanner_findings,
+        }),
+      })
+      if (!res.ok) throw new Error(await res.text())
+      setResult(await res.json())
+    } catch (e) {
+      setError(e instanceof Error ? e.message : 'Advisor fehlgeschlagen')
+    } finally { setLoading(false) }
+  }
+
+  // auto-recompute when certifications / target / scanner signals change (no button click needed)
+  useEffect(() => { if (certs.length) run() }, [certs, target, findings])  // eslint-disable-line react-hooks/exhaustive-deps
+
+  return (
+    <div className="min-h-screen bg-gray-50 py-8">
+      <div className="max-w-5xl mx-auto px-4">
+        <h1 className="text-3xl font-bold text-gray-900">ETO / Onboarding-Advisor</h1>
+        <p className="text-gray-600 mt-2 mb-6">
+          Zertifikate + Ziel + Scanner-Signale → Silent Pass → Capability-Delta + nächste beste Fragen.
+          Welt-1: ein Zertifikat <em>legt nahe</em>, beweist nichts (Verifikation erforderlich).
+        </p>
+
+        <div className="grid md:grid-cols-2 gap-4 mb-6">
+          <Section title="Unternehmen & Ziel">
+            <label className="block text-sm text-gray-600">Unternehmen
+              <input value={company} onChange={e => setCompany(e.target.value)} className="mt-1 w-full border rounded-lg px-3 py-2" /></label>
+            <label className="block text-sm text-gray-600 mt-3">Branche
+              <input value={industry} onChange={e => setIndustry(e.target.value)} className="mt-1 w-full border rounded-lg px-3 py-2" /></label>
+            <label className="block text-sm text-gray-600 mt-3">Ziel
+              <select value={target} onChange={e => setTarget(e.target.value)} className="mt-1 w-full border rounded-lg px-3 py-2">
+                {(targets.length ? targets : ['CRA']).map(t => <option key={t} value={t}>{t}</option>)}
+              </select></label>
+            <label className="block text-sm text-gray-600 mt-3">Vorhandene Nachweise (kommagetrennt)
+              <input value={knownEvidence} onChange={e => setKnownEvidence(e.target.value)} className="mt-1 w-full border rounded-lg px-3 py-2" /></label>
+          </Section>
+
+          <Section title="Zertifizierungen">
+            <div className="flex flex-wrap gap-2">
+              {CERTS.map(c => (
+                <button key={c} onClick={() => toggle(certs, setCerts, c)}
+                  className={`px-3 py-1.5 rounded-lg text-sm border ${certs.includes(c) ? 'bg-blue-600 text-white border-blue-600' : 'bg-white text-gray-700 border-gray-300'}`}>{c}</button>
+              ))}
+            </div>
+          </Section>
+        </div>
+
+        <Section title="Scanner-Signale (Silent Pass)" hint="observation = gesehen · partial = Indikation · requirement = gefordert (≠ vorhanden)">
+          <div className="grid sm:grid-cols-2 gap-2">
+            {FINDINGS.map(f => (
+              <label key={f.signal_id} className="flex items-center gap-2 text-sm text-gray-700">
+                <input type="checkbox" checked={findings.includes(f.signal_id)} onChange={() => toggle(findings, setFindings, f.signal_id)} />
+                <span>{f.label}</span>
+                <span className={`ml-auto text-[10px] px-1.5 py-0.5 rounded ${f.kind === 'requirement' ? 'bg-purple-100 text-purple-700' : f.kind === 'partial' ? 'bg-amber-100 text-amber-700' : 'bg-emerald-100 text-emerald-700'}`}>{f.kind}</span>
+              </label>
+            ))}
+          </div>
+        </Section>
+
+        <button onClick={run} disabled={loading || !certs.length}
+          className="mt-6 w-full py-3 bg-blue-600 text-white rounded-xl font-medium hover:bg-blue-700 disabled:opacity-50">
+          {loading ? 'Analysiere…' : 'Advisor starten'}
+        </button>
+
+        {error && <div className="mt-6 bg-red-50 border border-red-200 rounded-lg p-4 text-red-700 text-sm whitespace-pre-wrap">{error}</div>}
+
+        {result && (
+          <div className="mt-8 space-y-4">
+            <div className="bg-blue-600 text-white rounded-xl p-5">
+              <div className="text-lg font-semibold">{result.headline}</div>
+              <div className="text-blue-100 text-sm mt-1">{result.silent_intake_summary}</div>
+            </div>
+            <div className="grid md:grid-cols-2 gap-4">
+              <Section title="Automatisch erkannt" hint="konkrete Artefakte – nicht mehr gefragt"><Chips items={result.auto_detected.map(lbl)} tone="bg-emerald-100 text-emerald-800" /></Section>
+              <Section title="Indikationen" hint="erhöht Annahmestärke – trotzdem gefragt"><Chips items={result.indications.map(lbl)} tone="bg-amber-100 text-amber-800" /></Section>
+            </div>
+            <Section title="Nächste beste Fragen" hint="max 5, jede erklärt sich selbst">
+              {result.top_5_questions.length ? (
+                <ol className="space-y-3">
+                  {result.top_5_questions.map((q, i) => (
+                    <li key={q.capability_id} className="border-l-2 border-blue-300 pl-3">
+                      <div className="font-medium text-gray-900">{i + 1}. {lbl(q.capability_id)}</div>
+                      <div className="text-sm text-gray-600">{q.why}</div>
+                    </li>
+                  ))}
+                </ol>
+              ) : <span className="text-gray-400 text-sm">—</span>}
+            </Section>
+            <div className="grid md:grid-cols-2 gap-4">
+              <Section title="Wahrscheinlich abgedeckt (Welt-1)" hint="Zertifikat legt nahe – Verifikation erforderlich">
+                {result.inferred_assumptions.length ? result.inferred_assumptions.map(a => (
+                  <div key={a.certification} className="mb-2"><span className="font-medium">{a.certification}</span>: {a.capabilities.map(lbl).join(', ')}</div>
+                )) : <span className="text-gray-400 text-sm">—</span>}
+              </Section>
+              <Section title="Nicht relevant" hint="relevance(evidence, target) = 0">
+                {result.rejected_assumptions.length ? result.rejected_assumptions.map((a, i) => (
+                  <div key={i} className="mb-1 text-sm text-gray-700">{a.statement}</div>
+                )) : <span className="text-gray-400 text-sm">—</span>}
+              </Section>
+            </div>
+            <div className="grid md:grid-cols-2 gap-4">
+              <Section title="Offene Lücken (Delta)"><Chips items={result.capability_delta.map(lbl)} tone="bg-gray-100 text-gray-700" /></Section>
+              <Section title="Geforderte Nachweise"><Chips items={result.evidence_requests} tone="bg-gray-100 text-gray-700" /></Section>
+            </div>
+            <Section title="Vollständigkeit" hint={result.unsupported_domains.length ? `nicht abgedeckt: ${result.unsupported_domains.join(', ')}` : undefined}>
+              <span className="text-sm text-gray-700">{result.completeness_summary || '—'}</span>
+            </Section>
+          </div>
+        )}
+      </div>
+    </div>
+  )
+}

ai-compliance-sdk/internal/iace/proposer_pin.go

@@ -0,0 +1,73 @@
+package iace
+
+// P3: pin accepted proposer decisions into the GT gate.
+//
+// When a human accepts a proposal from the offline proposer (a dedup
+// supersession, a foreign-framing gate, a vocab→tag mapping, a coverage hazard),
+// they record an AcceptedPin. A pin is a tiny, machine-scoped invariant — "this
+// pattern MUST (or must NOT) fire for this machine" — that a test re-checks on
+// every run. This is what makes the library's growth COMPOUND into the gate
+// instead of silently eroding it: a future change that re-introduces a dropped
+// duplicate, un-gates a foreign pattern, or removes a coverage hazard breaks the
+// pin and fails CI.
+//
+// A single boolean covers all four proposal types:
+//   - dedup supersession accepted → DropPattern MustFire=false
+//   - foreign-framing gate accepted → foreign pattern MustFire=false
+//   - vocab→tag / coverage hazard accepted → the enabled pattern MustFire=true
+
+// AcceptedPin is one regression invariant for an accepted proposal.
+type AcceptedPin struct {
+	Pattern      string `json:"pattern"`
+	MustFire     bool   `json:"must_fire"`
+	Reason       string `json:"reason"`
+	FromProposal string `json:"from_proposal,omitempty"`
+}
+
+// PinSet is the accepted-pin registry for one machine (testdata/accepted_pins_*.json).
+type PinSet struct {
+	Machine string        `json:"machine"`
+	Pins    []AcceptedPin `json:"pins"`
+}
+
+// PinResult is the verdict for one pin against an engine run.
+type PinResult struct {
+	Pin    AcceptedPin
+	OK     bool
+	Detail string
+}
+
+// VerifyPins checks every pin against the set of pattern IDs the engine actually
+// fired for the machine. A pin holds iff the pattern's presence equals MustFire.
+func VerifyPins(pins []AcceptedPin, firedPatternIDs []string) []PinResult {
+	fired := make(map[string]bool, len(firedPatternIDs))
+	for _, id := range firedPatternIDs {
+		fired[id] = true
+	}
+	out := make([]PinResult, 0, len(pins))
+	for _, p := range pins {
+		got := fired[p.Pattern]
+		ok := got == p.MustFire
+		detail := "ok"
+		if !ok {
+			if p.MustFire {
+				detail = "expected to fire but did NOT — coverage/mapping regressed"
+			} else {
+				detail = "expected to be suppressed but FIRED — gate/supersession regressed"
+			}
+		}
+		out = append(out, PinResult{Pin: p, OK: ok, Detail: detail})
+	}
+	return out
+}
+
+// GenerateDedupPin turns an accepted (verdict=duplicate) dedup candidate into the
+// pin that protects the supersession: the dropped pattern must no longer fire.
+func GenerateDedupPin(c DedupCandidate) AcceptedPin {
+	return AcceptedPin{
+		Pattern:      c.DropPattern,
+		MustFire:     false,
+		Reason:       "accepted duplicate of " + c.KeepPattern + " (" + c.Category + ")",
+		FromProposal: "dedup " + c.DropPattern + " -> " + c.KeepPattern,
+	}
+}

ai-compliance-sdk/internal/iace/proposer_pin_test.go

@@ -0,0 +1,63 @@
+package iace
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestVerifyPins(t *testing.T) {
+	pins := []AcceptedPin{
+		{Pattern: "HPa", MustFire: true},
+		{Pattern: "HPb", MustFire: false},
+	}
+	res := VerifyPins(pins, []string{"HPa", "HPb"})
+	if !res[0].OK {
+		t.Errorf("HPa must_fire=true and it fired -> should be OK")
+	}
+	if res[1].OK {
+		t.Errorf("HPb must_fire=false but it fired -> should be VIOLATED")
+	}
+	res2 := VerifyPins(pins, []string{})
+	if res2[0].OK || !res2[1].OK {
+		t.Errorf("expected HPa violated + HPb ok, got %+v", res2)
+	}
+}
+
+func TestGenerateDedupPin(t *testing.T) {
+	pin := GenerateDedupPin(DedupCandidate{KeepPattern: "HP144", DropPattern: "HP013", Category: "electrical_hazard"})
+	if pin.Pattern != "HP013" || pin.MustFire {
+		t.Fatalf("want pin {HP013, must_fire=false}, got %+v", pin)
+	}
+}
+
+// TestWarewashing_AcceptedPins re-checks every accepted P1 supersession against the
+// live warewashing engine output. A future change that un-suppresses HP013/016/018
+// or drops HP2201/HP144 breaks a pin here — the gate compounds, not erodes.
+func TestWarewashing_AcceptedPins(t *testing.T) {
+	raw, err := os.ReadFile(filepath.Join("testdata", "accepted_pins_warewashing.json"))
+	if err != nil {
+		t.Fatalf("read pins: %v", err)
+	}
+	var ps PinSet
+	if err := json.Unmarshal(raw, &ps); err != nil {
+		t.Fatalf("parse pins: %v", err)
+	}
+
+	_, _, kept := warewashingEngineOutput()
+	firedIDs := make([]string, 0, len(kept))
+	for _, pm := range kept {
+		firedIDs = append(firedIDs, pm.PatternID)
+	}
+
+	ok := 0
+	for _, r := range VerifyPins(ps.Pins, firedIDs) {
+		if r.OK {
+			ok++
+			continue
+		}
+		t.Errorf("PIN VIOLATED: %s (must_fire=%v) — %s [%s]", r.Pin.Pattern, r.Pin.MustFire, r.Detail, r.Pin.Reason)
+	}
+	t.Logf("accepted pins for %q: %d/%d hold", ps.Machine, ok, len(ps.Pins))
+}

ai-compliance-sdk/internal/iace/testdata/accepted_pins_warewashing.json

@@ -0,0 +1,10 @@
+{
+  "machine": "Gewerbliche Untertisch-Geschirrspuelmaschine (vernetzt)",
+  "pins": [
+    {"pattern": "HP016", "must_fire": false, "reason": "generic hot-surface (Formwerkzeuge/Auspuffleitung framing) superseded by HP2201", "from_proposal": "P1 thermal supersession"},
+    {"pattern": "HP018", "must_fire": false, "reason": "actuator-burn superseded by HP2201", "from_proposal": "P1 thermal supersession"},
+    {"pattern": "HP013", "must_fire": false, "reason": "stored-energy Batterie/USV framing superseded by HP144", "from_proposal": "P1 stored-energy supersession"},
+    {"pattern": "HP2201", "must_fire": true, "reason": "warewashing hot-surface (Boiler/Tank/Spuelkammer) must remain — it is the clean equivalent that replaces HP016/HP018", "from_proposal": "P1 thermal supersession"},
+    {"pattern": "HP144", "must_fire": true, "reason": "residual-voltage (Frequenzumrichter/Zwischenkreis) must remain — clean equivalent that replaces HP013", "from_proposal": "P1 stored-energy supersession"}
+  ]
+}

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

@@ -78,6 +78,19 @@ func (c *LegalRAGClient) Search(ctx context.Context, query string, regulationIDs
 // If hybrid search is enabled, it uses the Qdrant Query API with RRF fusion
 // (dense + full-text). Falls back to dense-only /points/search on failure.
 func (c *LegalRAGClient) searchInternal(ctx context.Context, collection string, query string, regulationIDs []string, topK int) ([]LegalSearchResult, error) {
+	// Multi-Regulation-Retrieval: nennt die Query EXPLIZIT >=2 Regelwerke (z.B. "CRA und
+	// Maschinenverordnung"), wird pro Regelwerk separat retrieved + gemergt, damit BEIDE
+	// Domaenen im Prompt landen statt nur der keyword-dominanten. Generisch (Query->Regelwerke,
+	// keine doc-spezifische Logik); nur wenn der Caller nicht ohnehin schon auf Regulierungen
+	// filtert. Best-effort: leeres/fehlerhaftes Multi-Ergebnis faellt auf die Standardsuche zurueck.
+	if len(regulationIDs) == 0 {
+		if regs := detectRegulations(query); len(regs) >= 2 {
+			if mr, mErr := c.searchMultiRegulation(ctx, collection, query, regs, topK); mErr == nil && len(mr) > 0 {
+				return mr, nil
+			}
+		}
+	}
+
 	embedding, err := c.generateEmbedding(ctx, query)
 	if err != nil {
 		return nil, fmt.Errorf("failed to generate embedding: %w", err)
@@ -123,43 +136,7 @@ func (c *LegalRAGClient) searchInternal(ctx context.Context, collection string,
 		hits = c.expandViaGraph(ctx, collection, hits)
 	}
 
-	results := make([]LegalSearchResult, len(hits))
-	for i, hit := range hits {
-		// Legal-Metadaten nach rag_reingest_spec.md §2: bevorzugt die normalisierten Felder
-		// (article_label/regulation_code/article/...); Fallback auf alte Feldnamen, solange der
-		// Korpus noch nicht re-ingestiert ist (regulation_id, section="§ 38").
-		regCode := getString(hit.Payload, "regulation_code")
-		if regCode == "" {
-			regCode = getString(hit.Payload, "regulation_id")
-		}
-		article := getString(hit.Payload, "article")
-		if article == "" {
-			article = getString(hit.Payload, "section")
-		}
-		results[i] = LegalSearchResult{
-			Text:            getString(hit.Payload, "chunk_text"),
-			RegulationCode:  regCode,
-			RegulationName:  getString(hit.Payload, "regulation_name_de"),
-			RegulationShort: getString(hit.Payload, "regulation_short"),
-			Category:        getString(hit.Payload, "category"),
-			ArticleLabel:    getString(hit.Payload, "article_label"),
-			Article:         article,
-			Paragraph:       getString(hit.Payload, "paragraph"),
-			Sub:             getString(hit.Payload, "sub"),
-			IsRecital:       getBool(hit.Payload, "is_recital"),
-			CitationStyle:   getString(hit.Payload, "citation_style"),
-			Pages:           getIntSlice(hit.Payload, "pages"),
-			SourceURL:       getString(hit.Payload, "source"),
-			Score:           hit.Score,
-			AuthorityWeight: getInt(hit.Payload, "authority_weight"),
-			SourceClass:     getString(hit.Payload, "source_class"),
-			Jurisdiction:    getString(hit.Payload, "jurisdiction"),
-			CitationUnit:    getString(hit.Payload, "citation_unit"),
-			ReferencesOut:   getStringSlice(hit.Payload, "references_out"),
-			ReferencesIn:    getStringSlice(hit.Payload, "references_in"),
-			Superseded:      getString(hit.Payload, "status") == "superseded",
-		}
-	}
+	results := hitsToResults(hits)
 
 	// Authority-aware Re-Ranking: bindendes Recht der passenden Jurisdiktion/Domaene nach
 	// oben, Guidance/Fremdrecht/Off-Domain runter (nichts wird geloescht). Reihenfolge only,

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

@@ -122,12 +122,14 @@ func (c *LegalRAGClient) searchHybrid(ctx context.Context, collection string, em
 	}
 
 	if len(regulationIDs) > 0 {
-		conditions := make([]qdrantCondition, len(regulationIDs))
-		for i, regID := range regulationIDs {
-			conditions[i] = qdrantCondition{
-				Key:   "regulation_id",
-				Match: qdrantMatch{Value: regID},
-			}
+		// Match BOTH the legacy field (regulation_id) and the normalized field
+		// (regulation_code) so per-regulation filtering works on the re-ingested corpus too.
+		conditions := make([]qdrantCondition, 0, len(regulationIDs)*2)
+		for _, regID := range regulationIDs {
+			conditions = append(conditions,
+				qdrantCondition{Key: "regulation_id", Match: qdrantMatch{Value: regID}},
+				qdrantCondition{Key: "regulation_code", Match: qdrantMatch{Value: regID}},
+			)
 		}
 		queryReq.Filter = &qdrantFilter{Should: conditions}
 	}
@@ -175,12 +177,14 @@ func (c *LegalRAGClient) searchDense(ctx context.Context, collection string, emb
 	}
 
 	if len(regulationIDs) > 0 {
-		conditions := make([]qdrantCondition, len(regulationIDs))
-		for i, regID := range regulationIDs {
-			conditions[i] = qdrantCondition{
-				Key:   "regulation_id",
-				Match: qdrantMatch{Value: regID},
-			}
+		// Match BOTH the legacy field (regulation_id) and the normalized field
+		// (regulation_code) so per-regulation filtering works on the re-ingested corpus too.
+		conditions := make([]qdrantCondition, 0, len(regulationIDs)*2)
+		for _, regID := range regulationIDs {
+			conditions = append(conditions,
+				qdrantCondition{Key: "regulation_id", Match: qdrantMatch{Value: regID}},
+				qdrantCondition{Key: "regulation_code", Match: qdrantMatch{Value: regID}},
+			)
 		}
 		searchReq.Filter = &qdrantFilter{Should: conditions}
 	}

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

@@ -0,0 +1,143 @@
+package ucca
+
+import (
+	"context"
+	"fmt"
+	"strings"
+)
+
+// multiRegMinPerRegulation is the minimum number of hits fetched per named regulation, so
+// each domain is fairly represented even when topK/len(regs) would be tiny.
+const multiRegMinPerRegulation = 3
+
+// regulationCatalog maps a regulation to (a) the aliases that signal it is EXPLICITLY named
+// in a query and (b) the regulation_code/regulation_id values used to filter the corpus.
+// Deterministic + generic: a query naming >=2 regulations triggers per-regulation retrieval
+// so a cross-regulation question returns every named domain — NOT a doc-specific rule.
+var regulationCatalog = []struct {
+	Canonical  string
+	Aliases    []string
+	CodeValues []string
+}{
+	{"CRA", []string{"cra", "cyber resilience"}, []string{"CRA"}},
+	{"MaschVO", []string{"maschinenverordnung", "maschvo", "machinery regulation"}, []string{"MASCHVO", "MaschVO"}},
+	{"NIS2", []string{"nis2", "nis-2", "nis 2"}, []string{"NIS2"}},
+	{"DORA", []string{"dora"}, []string{"DORA"}},
+	{"Data Act", []string{"data act", "datengesetz"}, []string{"DATA ACT", "DataAct"}},
+	{"AI Act", []string{"ai act", "ki-vo", "ki-verordnung", "ai-verordnung"}, []string{"AI ACT", "AIAct"}},
+	{"DSGVO", []string{"dsgvo", "gdpr"}, []string{"DSGVO"}},
+	{"TDDDG", []string{"tdddg"}, []string{"TDDDG"}},
+	{"BDSG", []string{"bdsg"}, []string{"BDSG"}},
+}
+
+type detectedRegulation struct {
+	Canonical  string
+	CodeValues []string
+}
+
+// detectRegulations returns the DISTINCT regulations explicitly named in the query. >=2 of
+// them is the trigger for multi-regulation retrieval. Pure + deterministic, no LLM.
+func detectRegulations(query string) []detectedRegulation {
+	q := strings.ToLower(query)
+	var out []detectedRegulation
+	for _, r := range regulationCatalog {
+		for _, a := range r.Aliases {
+			if strings.Contains(q, a) {
+				out = append(out, detectedRegulation{Canonical: r.Canonical, CodeValues: r.CodeValues})
+				break
+			}
+		}
+	}
+	return out
+}
+
+func hitID(h qdrantSearchHit) string { return fmt.Sprintf("%v", h.ID) }
+
+// searchMultiRegulation retrieves each explicitly-named regulation SEPARATELY (per-regulation
+// filter) and merges, so a cross-regulation query ("Wie greifen CRA und MaschVO ineinander?")
+// returns BOTH domains in the prompt instead of only the keyword-dominant one. Generic over any
+// named pair (DSGVO+TDDDG, CRA+NIS2, DORA+NIS2, AI Act+DSGVO, ...). The merged pool is
+// authority-reranked once. Pure pool-construction; topK contract preserved.
+func (c *LegalRAGClient) searchMultiRegulation(ctx context.Context, collection, query string, regs []detectedRegulation, topK int) ([]LegalSearchResult, error) {
+	embedding, err := c.generateEmbedding(ctx, query)
+	if err != nil {
+		return nil, fmt.Errorf("failed to generate embedding: %w", err)
+	}
+	perReg := topK / len(regs)
+	if perReg < multiRegMinPerRegulation {
+		perReg = multiRegMinPerRegulation
+	}
+	var merged []qdrantSearchHit
+	seen := make(map[string]bool)
+	for _, r := range regs {
+		var hits []qdrantSearchHit
+		if c.hybridEnabled {
+			if h, hErr := c.searchHybrid(ctx, collection, embedding, r.CodeValues, perReg); hErr == nil {
+				hits = h
+			}
+		}
+		if hits == nil {
+			if h, dErr := c.searchDense(ctx, collection, embedding, r.CodeValues, perReg); dErr == nil {
+				hits = h
+			}
+		}
+		for _, h := range hits {
+			id := hitID(h)
+			if seen[id] {
+				continue
+			}
+			seen[id] = true
+			merged = append(merged, h)
+		}
+	}
+	if len(merged) == 0 {
+		return nil, fmt.Errorf("multi-regulation search returned no hits")
+	}
+	results := hitsToResults(merged)
+	results = rerankByAuthority(query, results)
+	if topK > 0 && len(results) > topK {
+		results = results[:topK]
+	}
+	return results, nil
+}
+
+// hitsToResults maps raw Qdrant hits to LegalSearchResult, preferring the normalized payload
+// fields (regulation_code/article_label/...) with fallback to the legacy names (regulation_id,
+// section) while the corpus is mid-re-ingestion. Shared by searchInternal + searchMultiRegulation.
+func hitsToResults(hits []qdrantSearchHit) []LegalSearchResult {
+	results := make([]LegalSearchResult, len(hits))
+	for i, hit := range hits {
+		regCode := getString(hit.Payload, "regulation_code")
+		if regCode == "" {
+			regCode = getString(hit.Payload, "regulation_id")
+		}
+		article := getString(hit.Payload, "article")
+		if article == "" {
+			article = getString(hit.Payload, "section")
+		}
+		results[i] = LegalSearchResult{
+			Text:            getString(hit.Payload, "chunk_text"),
+			RegulationCode:  regCode,
+			RegulationName:  getString(hit.Payload, "regulation_name_de"),
+			RegulationShort: getString(hit.Payload, "regulation_short"),
+			Category:        getString(hit.Payload, "category"),
+			ArticleLabel:    getString(hit.Payload, "article_label"),
+			Article:         article,
+			Paragraph:       getString(hit.Payload, "paragraph"),
+			Sub:             getString(hit.Payload, "sub"),
+			IsRecital:       getBool(hit.Payload, "is_recital"),
+			CitationStyle:   getString(hit.Payload, "citation_style"),
+			Pages:           getIntSlice(hit.Payload, "pages"),
+			SourceURL:       getString(hit.Payload, "source"),
+			Score:           hit.Score,
+			AuthorityWeight: getInt(hit.Payload, "authority_weight"),
+			SourceClass:     getString(hit.Payload, "source_class"),
+			Jurisdiction:    getString(hit.Payload, "jurisdiction"),
+			CitationUnit:    getString(hit.Payload, "citation_unit"),
+			ReferencesOut:   getStringSlice(hit.Payload, "references_out"),
+			ReferencesIn:    getStringSlice(hit.Payload, "references_in"),
+			Superseded:      getString(hit.Payload, "status") == "superseded",
+		}
+	}
+	return results
+}

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

@@ -0,0 +1,92 @@
+package ucca
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+	"testing"
+)
+
+// TestDetectRegulations is a pure unit test of the multi-regulation TRIGGER (no Qdrant):
+// only an explicit naming of >=2 regulations enables multi-regulation retrieval. A single
+// named regulation, or a topical question that doesn't name one, stays single-domain.
+func TestDetectRegulations(t *testing.T) {
+	cases := []struct {
+		q    string
+		want int
+	}{
+		{"Welche neun Kriterien nennt WP248 fuer ein voraussichtlich hohes Risiko?", 0},
+		{"Welche Anforderungen gelten fuer wesentliche Veraenderungen einer Maschine?", 0}, // "Maschine" != MaschVO
+		{"Benoetigt eine SPS ohne Netzwerkanschluss eine CRA-Bewertung?", 1},               // 1 -> single
+		{"Wie greifen CRA und Maschinenverordnung bei einer vernetzten Maschine ineinander?", 2},
+		{"Wie greifen DSGVO und TDDDG bei der Nutzung von Cookies ineinander?", 2},
+		{"Wie verhalten sich DORA und NIS2 fuer ein Finanzunternehmen?", 2},
+		{"Wie greifen AI Act und DSGVO bei einem KI-System ineinander?", 2},
+	}
+	for _, c := range cases {
+		if got := len(detectRegulations(c.q)); got != c.want {
+			t.Errorf("detectRegulations(%q) = %d, want %d", c.q, got, c.want)
+		}
+	}
+}
+
+// TestMultiRegE2E (RUN_E2E=1) verifies against the build collection that an explicit
+// cross-regulation query returns BOTH named domains in the top-K — the core acceptance
+// gate for multi-regulation retrieval.
+func TestMultiRegE2E(t *testing.T) {
+	if os.Getenv("RUN_E2E") != "1" {
+		t.Skip("set RUN_E2E=1 + QDRANT_URL/OLLAMA_URL")
+	}
+	c := NewLegalRAGClient()
+	coll := os.Getenv("E2E_COLLECTION")
+	if coll == "" {
+		coll = "bp_compliance_kb_2026_1_build"
+	}
+	cases := []struct {
+		id   string
+		q    string
+		want []string
+	}{
+		{"GQ-0070 CRA+MaschVO", "Wie greifen CRA und Maschinenverordnung bei einer vernetzten Maschine ineinander?", []string{"CRA", "MASCH"}},
+		{"DSGVO+TDDDG", "Wie greifen DSGVO und TDDDG bei der Nutzung von Cookies und Tracking-Technologien ineinander?", []string{"DSGVO", "TDDDG"}},
+		{"CRA+NIS2", "Wie verhalten sich CRA und NIS2 bei einem vernetzten Produkt eines wichtigen Unternehmens zueinander?", []string{"CRA", "NIS2"}},
+		{"DORA+NIS2", "Wie greifen DORA und NIS2 bei einem Finanzunternehmen ineinander?", []string{"DORA", "NIS2"}},
+		{"AI Act+DSGVO", "Wie greifen AI Act und DSGVO bei einem KI-System ineinander, das personenbezogene Daten verarbeitet?", []string{"AI ACT", "DSGVO"}},
+	}
+	for _, tc := range cases {
+		res, err := c.SearchCollection(context.Background(), coll, tc.q, nil, 8)
+		if err != nil {
+			t.Fatalf("%s: %v", tc.id, err)
+		}
+		present := map[string]bool{}
+		for _, r := range res {
+			present[strings.ToUpper(r.RegulationCode)] = true
+		}
+		ok := true
+		for _, w := range tc.want {
+			found := false
+			for cd := range present {
+				if strings.Contains(cd, w) {
+					found = true
+					break
+				}
+			}
+			if !found {
+				ok = false
+			}
+		}
+		codes := make([]string, 0, len(present))
+		for cd := range present {
+			codes = append(codes, cd)
+		}
+		status := "OK"
+		if !ok {
+			status = "FAIL"
+		}
+		fmt.Printf("%-22s want=%v  present=%v  %s\n", tc.id, tc.want, codes, status)
+		if !ok {
+			t.Errorf("%s: not all named regulations in top-8 (want %v, got %v)", tc.id, tc.want, codes)
+		}
+	}
+}

backend-compliance/compliance/api/onboarding_routes.py

@@ -8,7 +8,7 @@ This adds NO new reasoning logic. It exposes the already-built, tested orchestra
 """
 
 import logging
-from typing import List, Optional
+from typing import Dict, List, Optional
 
 from fastapi import APIRouter, HTTPException
 from pydantic import BaseModel, Field
@@ -20,7 +20,7 @@ from compliance.onboarding import (
     ProducedSignal,
     RejectedAssumption,
 )
-from compliance.services.onboarding_service import run_advisor, supported_targets
+from compliance.services.onboarding_service import labels_for, run_advisor, supported_targets
 
 logger = logging.getLogger(__name__)
 router = APIRouter(prefix="/onboarding", tags=["onboarding"])
@@ -50,6 +50,7 @@ class AdvisorResponse(BaseModel):
     evidence_requests: List[str] = Field(default_factory=list)
     unsupported_domains: List[str] = Field(default_factory=list)
     completeness_summary: str = ""
+    capability_labels: Dict[str, str] = Field(default_factory=dict)   # capability_id -> human label (DE)
 
 
 @router.get("/targets")
@@ -65,10 +66,17 @@ def advisor_start_endpoint(req: OnboardingAdvisorRequest) -> AdvisorResponse:
         company=req.company, certifications=req.certifications, target=req.target,
         signals=req.scanner_findings, known_evidence=req.known_evidence,
         products=req.products, markets=req.markets, industry=req.industry or "")
+    surfaced = [
+        *result.auto_detected, *result.indications, *result.capability_delta,
+        *(q.capability_id for q in result.next_best_questions),
+        *(c for a in result.inferred_assumptions for c in a.capabilities),
+        *(m.capability_id for m in result.top_measures),
+    ]
     return AdvisorResponse(
         silent_intake_summary=si_summary, headline=result.headline, auto_detected=result.auto_detected,
         indications=result.indications,
         inferred_assumptions=result.inferred_assumptions, rejected_assumptions=result.rejected_assumptions,
         top_5_questions=result.next_best_questions, capability_delta=result.capability_delta,
         top_measures=result.top_measures, evidence_requests=result.evidence_requests,
-        unsupported_domains=result.unsupported_domains, completeness_summary=result.completeness_summary)
+        unsupported_domains=result.unsupported_domains, completeness_summary=result.completeness_summary,
+        capability_labels=labels_for(surfaced))

backend-compliance/compliance/onboarding/__init__.py

@@ -21,6 +21,14 @@ from .observations import (
     empirical_distribution,
     reviewed,
 )
+from .observation_log import (
+    HypothesisStats,
+    ObservationRecord,
+    aggregate_by_hypothesis,
+    append_observation,
+    load_observations,
+    review_queue,
+)
 from .signals import (
     ProducedSignal,
     SignalVocabularyEntry,
@@ -69,4 +77,10 @@ __all__ = [
     "ProducedSignal",
     "SignalVocabularyEntry",
     "normalize_signals",
+    "ObservationRecord",
+    "HypothesisStats",
+    "append_observation",
+    "load_observations",
+    "aggregate_by_hypothesis",
+    "review_queue",
 ]

backend-compliance/compliance/onboarding/engine.py

@@ -143,8 +143,8 @@ def advisor_start(
         next_best_questions=next_q, capability_delta=delta, top_measures=measures,
         evidence_requests=evidence, unsupported_domains=unsupported,
         completeness_summary=rep.completeness_summary,
-        headline="%d Anforderungen erkannt · %d automatisch erkannt (Intake) · %d wahrscheinlich (Zertifikate) · %d zu klären"
-        % (len(assess.coverage), len(auto_detected), len(probably), len(next_q)))
+        headline="%d von %d Anforderungen offen · %d automatisch erkannt (Intake) · %d wahrscheinlich (Zertifikate) · %d zu klären"
+        % (len(delta), len(assess.coverage), len(auto_detected), len(probably), len(next_q)))
 
 
 def apply_answer(known_capabilities: Sequence[str], capability_id: str, answer: str) -> List[str]:

backend-compliance/compliance/onboarding/observation_log.py

@@ -0,0 +1,108 @@
+"""Observation Log — append-only JSONL store for empirical calibration events (Task 59b v1).
+
+Observations are NOT business data and NOT product-DB data — they are CALIBRATION events for the
+knowledge base ("ISO27001 -> SDL confirmed", "TISAX -> supplier security refuted"). So they live with the
+other versioned knowledge artifacts (hypotheses, transition patterns, vocabulary), NOT in the product
+database: an append-only JSONL log under `knowledge/observations/`. NO migration, NO DB. The empirical
+DISTRIBUTION and CONFIDENCE are COMPUTED from this log on demand (computed-not-stored) — a hypothesis is
+NEVER auto-updated; only REVIEWED observations calibrate (the review gate, enforced in observations.py).
+
+Append-only: each line is one ObservationRecord and lines are NEVER modified in place. A later review is
+a NEW line with the same observation_id and reviewed=true; load_observations() reconciles to the latest
+per id. You can `rm` the log and recompute, `git diff` it over months, or rebuild confidence under a new
+policy. Anonymisation is MANDATORY: customer_archetype is a sector/cert archetype, NEVER a real company
+name (this file is committed to git). Time is stamped by the CALLER (no hidden clock) for determinism.
+I/O only at the append/load boundary; statistics are pure. Python 3.9 compatible.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from typing import Dict, List, Optional, Sequence
+
+from pydantic import BaseModel, Field
+
+from .observations import Observation, empirical_confidence, empirical_distribution
+
+_DEFAULT_LOG = os.path.join(
+    os.path.dirname(__file__), "..", "..", "knowledge", "observations", "observations.jsonl")
+
+
+class ObservationRecord(Observation):
+    """A persisted observation line: an Observation (with its review gate + observation_type) plus log
+    metadata. `observation_id` is stable — a review re-appends the SAME id with reviewed=true."""
+
+    observation_id: str                                  # stable id; a review re-appends the same id
+    timestamp: str = ""                                  # ISO 8601, stamped by the CALLER (no hidden clock)
+    customer_archetype: str = ""                         # sector/cert archetype — NEVER a real company name
+    evidence: str = ""                                   # what backs the answer (reference, not the artifact)
+    provenance: str = ""                                 # where the answer came from (audit trail)
+    knowledge_version: str = ""                          # hypotheses/vocabulary version observed under
+
+
+class HypothesisStats(BaseModel):
+    """Per-hypothesis empirical rollup — all COMPUTED from the log, nothing stored on the hypothesis."""
+
+    hypothesis_id: str
+    distribution: Dict[str, int] = Field(default_factory=dict)   # reviewed counts per observation_type
+    confidence: Optional[float] = None                           # None until a for/against obs is reviewed
+    reviewed_count: int = 0
+    total_count: int = 0
+
+
+def append_observation(record: ObservationRecord, path: str = _DEFAULT_LOG) -> None:
+    """Append ONE record as a JSON line. Append-only — existing lines are never rewritten."""
+    os.makedirs(os.path.dirname(path), exist_ok=True)
+    line = json.dumps(record.model_dump(mode="json"), ensure_ascii=False, sort_keys=True)
+    with open(path, "a", encoding="utf-8") as fh:
+        fh.write(line + "\n")
+
+
+def load_observations(path: str = _DEFAULT_LOG, reconcile: bool = True) -> List[ObservationRecord]:
+    """Read all records — a single `.jsonl` file or a directory of monthly `.jsonl` files. With
+    reconcile, the LATEST record per observation_id wins (a later reviewed=true supersedes the original).
+    Returns deterministic order (by observation_id when reconciled, else append order)."""
+    files: List[str] = []
+    if os.path.isdir(path):
+        files = sorted(os.path.join(path, f) for f in os.listdir(path) if f.endswith(".jsonl"))
+    elif os.path.exists(path):
+        files = [path]
+    records: List[ObservationRecord] = []
+    for fpath in files:
+        with open(fpath, encoding="utf-8") as fh:
+            for raw in fh:
+                raw = raw.strip()
+                if raw:
+                    records.append(ObservationRecord(**json.loads(raw)))
+    if not reconcile:
+        return records
+    latest: Dict[str, ObservationRecord] = {}
+    for r in records:                                    # file/append order -> later lines win
+        latest[r.observation_id] = r
+    return [latest[k] for k in sorted(latest)]
+
+
+def aggregate_by_hypothesis(records: Sequence[ObservationRecord]) -> List[HypothesisStats]:
+    """Per-hypothesis distribution + confidence. The review gate applies inside empirical_distribution/
+    empirical_confidence (reviewed-only), so unreviewed observations are counted in total but never
+    calibrate. Deterministic order (by hypothesis id)."""
+    by_hyp: Dict[str, List[ObservationRecord]] = {}
+    for r in records:
+        by_hyp.setdefault(r.hypothesis_id, []).append(r)
+    out: List[HypothesisStats] = []
+    for hyp in sorted(by_hyp):
+        obs = by_hyp[hyp]
+        out.append(HypothesisStats(
+            hypothesis_id=hyp,
+            distribution=empirical_distribution(obs),    # reviewed-only (the gate)
+            confidence=empirical_confidence(obs),        # None until reviewed for/against
+            reviewed_count=sum(1 for o in obs if o.reviewed),
+            total_count=len(obs)))
+    return out
+
+
+def review_queue(records: Sequence[ObservationRecord]) -> List[ObservationRecord]:
+    """The reviewer's worklist: observations not yet reviewed. Calibration ignores these until a reviewer
+    accepts them (Observation -> Review -> Accepted -> Knowledge recomputed), never Observation -> conf++."""
+    return [r for r in records if not r.reviewed]

backend-compliance/compliance/services/onboarding_service.py

@@ -9,7 +9,7 @@ It adds NO new reasoning logic — it only exposes what exists. No DB, no persis
 from __future__ import annotations
 
 import os
-from typing import Any, Dict, List, Sequence, Tuple
+from typing import Any, Dict, Iterable, List, Sequence, Tuple
 
 import yaml
 
@@ -37,6 +37,13 @@ def _load(*parts: str) -> Any:
 _HYP_LIB = [CapabilityHypothesis(**h) for h in _load("certification_hypotheses", "hypotheses.yaml")["hypotheses"]]
 _VOCAB = [SignalVocabularyEntry(**v) for v in _load("onboarding", "signal_vocabulary.yaml")["signals"]]
 _SIGNAL_MAP = [SignalMapping(**m) for m in _load("onboarding", "intake_signal_map.yaml")["mappings"]]
+_LABELS: Dict[str, str] = _load("onboarding", "capability_labels.yaml")["labels"]
+
+
+def labels_for(capability_ids: Iterable[str]) -> Dict[str, str]:
+    """Human labels (DE) for the given capability ids — presentation only. Ids without a curated label
+    are omitted (the frontend falls back to a prettified id). Deduped, deterministic."""
+    return {c: _LABELS[c] for c in dict.fromkeys(capability_ids) if c in _LABELS}
 
 # target id -> transition pattern that defines its required capabilities (curated registry)
 _TARGET_PATTERNS = {
@@ -53,9 +60,10 @@ def supported_targets() -> List[str]:
 
 def _target(target_id: str) -> Tuple[List[TargetRequirement], Dict[str, List[str]]]:
     pat = _load("transition_patterns", _TARGET_PATTERNS[target_id])
-    reqs = [TargetRequirement(capability_id=a["capability"]) for a in pat["likely_covered"]]
+    reqs = [TargetRequirement(capability_id=a["capability"], rationale=a.get("reviewable_claim", "")) for a in pat["likely_covered"]]
     reqs += [TargetRequirement(capability_id=d["capability"], question_intent=d.get("needed_information", "verify_existence"),
-                               expected_evidence=d.get("expected_evidence", [])) for d in pat["delta_requirements"]]
+                               rationale=d.get("why_asked", ""), expected_evidence=d.get("expected_evidence", []))
+             for d in pat["delta_requirements"]]
     covers = {d["capability"]: d.get("covers_targets", []) for d in pat["delta_requirements"]}
     return reqs, covers
 

backend-compliance/compliance/transition_reasoning/engine.py

@@ -104,7 +104,8 @@ def assess_transition(
         )
         buckets[status].append(req.capability_id)
         if status in _REQUESTABLE:
-            reason, prio = _REQUESTABLE[status]
+            default_reason, prio = _REQUESTABLE[status]
+            reason = req.rationale or default_reason   # curated human text wins over the generic fallback
             requests.append(
                 TransitionQuestionRequest(
                     capability_id=req.capability_id,

backend-compliance/compliance/transition_reasoning/schemas.py

@@ -70,6 +70,7 @@ class TargetRequirement(BaseModel):
 
     capability_id: str  # MCAP-...
     question_intent: str = "verify_existence"  # passed through to the request, not rendered
+    rationale: str = ""  # curated human text (e.g. why_asked / reviewable_claim) — surfaced as the request reason
     expected_evidence: List[str] = Field(default_factory=list)
     source_control_id: Optional[str] = None
     supports_obligations: List[str] = Field(default_factory=list)

backend-compliance/knowledge/observations/.gitkeep

@@ -0,0 +1,2 @@
+# Append-only observation log (Task 59b). Real lines (observations.jsonl / YYYY-MM.jsonl) are written at
+# runtime via compliance/onboarding/observation_log.py. Anonymised archetypes only — NEVER real company names.

backend-compliance/knowledge/onboarding/capability_labels.yaml

@@ -0,0 +1,45 @@
+# Human-readable capability labels (DE) — presentation only, reusable across all targets.
+# A capability id is the stable machine identity; this maps it to an expert-facing label for the UI.
+# Curated knowledge (draft — to be corrected by the domain expert). Missing ids fall back to a
+# prettified id in the frontend. NO real company names. Keep labels short + concrete.
+
+labels:
+  # ── ISMS / ISO 27001 core ───────────────────────────────────────────────
+  information_security_management: "Informationssicherheits-Managementsystem (ISMS)"
+  access_control_and_authentication: "Zugriffskontrolle & Authentifizierung"
+  asset_and_configuration_management: "Asset- & Konfigurationsverwaltung"
+  cryptography: "Kryptographie / Verschlüsselung"
+  incident_management: "Security-Incident-Management"
+  security_awareness_training: "Security-Awareness-Schulungen"
+  supplier_security: "Lieferanten-Sicherheit"
+  security_logging_and_monitoring: "Security-Logging & Monitoring"
+  technical_vulnerability_management: "Technisches Schwachstellen-Management"
+  # ── TISAX / VDA-spezifisch ──────────────────────────────────────────────
+  prototype_protection: "Prototypenschutz (physisch & logisch)"
+  tisax_label_scope_selection: "TISAX-Label-/Scope-Festlegung"
+  tisax_assessment_via_enx: "TISAX-Assessment über die ENX-Plattform"
+  vda_isa_self_assessment: "VDA-ISA-Selbstauskunft"
+  data_protection_processing_on_behalf: "Auftragsverarbeitung (Art. 28 DSGVO)"
+  physical_security: "Physische Sicherheit / Zutrittskontrolle"
+  # ── QM / ISO 9001 ───────────────────────────────────────────────────────
+  document_and_change_control: "Dokumenten- & Änderungslenkung"
+  supplier_evaluation: "Lieferantenbewertung"
+  release_and_approval_process: "Freigabe- & Genehmigungsprozess"
+  ce_conformity_assessment_and_technical_documentation: "CE-Konformitätsbewertung & technische Dokumentation"
+  # ── CRA / Produkt-Cybersecurity ─────────────────────────────────────────
+  sbom_creation: "SBOM-Erstellung (Software-Stückliste)"
+  coordinated_vulnerability_disclosure: "Coordinated Vulnerability Disclosure (CVD)"
+  secure_development_lifecycle: "Sicherer Entwicklungslebenszyklus (SDLC)"
+  secure_signed_update_distribution: "Sichere, signierte Update-Verteilung"
+  security_update_support_period: "Sicherheits-Update-Supportzeitraum"
+  product_cyber_risk_assessment: "Produkt-Cyber-Risikobewertung"
+  exploited_vuln_and_incident_reporting: "Meldung ausgenutzter Schwachstellen & Vorfälle"
+  public_security_advisories: "Öffentliche Security Advisories"
+  cybersecurity_management_system: "Cybersecurity-Managementsystem (CSMS)"
+  # ── MaschinenVO / Safety ────────────────────────────────────────────────
+  machine_safety_risk_assessment: "Maschinen-Risikobeurteilung"
+  mechanical_safety_and_guards: "Mechanische Sicherheit & Schutzeinrichtungen"
+  operating_instructions_and_safety_information: "Betriebsanleitung & Sicherheitshinweise"
+  protection_against_corruption_of_safety_functions: "Schutz der Sicherheitsfunktionen vor Manipulation"
+  # ── Umwelt ──────────────────────────────────────────────────────────────
+  environmental_management_documentation: "Umweltmanagement-Dokumentation"

backend-compliance/tests/test_observation_log.py

@@ -0,0 +1,73 @@
+"""Observation Log — append-only JSONL store + computed statistics (Task 59b/c v1).
+
+Pins the user's decision (2026-06-28): observations are CALIBRATION data, not product data -> an
+append-only JSONL log under knowledge/observations/, NO DB, NO migration. Distribution and confidence are
+COMPUTED from the log; only REVIEWED observations calibrate (review gate); a later review is a new line
+that supersedes by observation_id. Nothing is ever written back to a hypothesis.
+"""
+
+from __future__ import annotations
+
+from compliance.onboarding import (
+    ObservationRecord,
+    ObservationType,
+    aggregate_by_hypothesis,
+    append_observation,
+    load_observations,
+    review_queue,
+)
+
+
+def _rec(oid, hyp, otype, reviewed=False, **kw):
+    return ObservationRecord(
+        observation_id=oid, hypothesis_id=hyp, observation_type=otype, reviewed=reviewed,
+        timestamp="2026-07-01T00:00:00Z", customer_archetype="machine_builder+ISO27001", **kw)
+
+
+def test_append_only_round_trip(tmp_path):
+    p = str(tmp_path / "obs.jsonl")
+    append_observation(_rec("o1", "HYP-secure_dev", ObservationType.CONFIRMED, reviewed=True), p)
+    append_observation(_rec("o2", "HYP-secure_dev", ObservationType.REFUTED, reviewed=True), p)
+    recs = load_observations(p)
+    assert {r.observation_id for r in recs} == {"o1", "o2"}
+    assert all(r.customer_archetype == "machine_builder+ISO27001" for r in recs)  # anonymised archetype, not a name
+
+
+def test_review_supersedes_by_id_append_only(tmp_path):
+    p = str(tmp_path / "obs.jsonl")
+    append_observation(_rec("o1", "HYP-x", ObservationType.CONFIRMED, reviewed=False), p)   # raw answer
+    append_observation(_rec("o1", "HYP-x", ObservationType.CONFIRMED, reviewed=True,
+                            reviewed_by="anna"), p)                                          # later review event
+    assert len(load_observations(p, reconcile=False)) == 2                                  # both lines kept (append-only)
+    recs = load_observations(p)                                                             # reconciled
+    assert len(recs) == 1 and recs[0].reviewed and recs[0].reviewed_by == "anna"
+
+
+def test_statistics_apply_the_review_gate(tmp_path):
+    p = str(tmp_path / "obs.jsonl")
+    append_observation(_rec("a", "HYP-sdl", ObservationType.CONFIRMED, reviewed=True), p)
+    append_observation(_rec("b", "HYP-sdl", ObservationType.CONFIRMED, reviewed=True), p)
+    append_observation(_rec("c", "HYP-sdl", ObservationType.REFUTED, reviewed=True), p)
+    append_observation(_rec("d", "HYP-sdl", ObservationType.CONFIRMED, reviewed=False), p)  # unreviewed -> ignored
+    stats = {s.hypothesis_id: s for s in aggregate_by_hypothesis(load_observations(p))}
+    s = stats["HYP-sdl"]
+    assert s.total_count == 4 and s.reviewed_count == 3
+    assert s.distribution["confirmed"] == 2 and s.distribution["refuted"] == 1   # unreviewed one excluded
+    assert s.confidence == round(2 / 3, 2)                                        # (2 + 0.5*0) / 3
+
+
+def test_review_queue_lists_unreviewed(tmp_path):
+    p = str(tmp_path / "obs.jsonl")
+    append_observation(_rec("a", "HYP-y", ObservationType.CONFIRMED, reviewed=True), p)
+    append_observation(_rec("b", "HYP-y", ObservationType.PARTIAL, reviewed=False), p)
+    q = review_queue(load_observations(p))
+    assert [r.observation_id for r in q] == ["b"]
+
+
+def test_load_directory_of_monthly_files(tmp_path):
+    d = tmp_path / "observations"
+    d.mkdir()
+    append_observation(_rec("a", "HYP-z", ObservationType.CONFIRMED, reviewed=True), str(d / "2026-06.jsonl"))
+    append_observation(_rec("b", "HYP-z", ObservationType.REFUTED, reviewed=True), str(d / "2026-07.jsonl"))
+    recs = load_observations(str(d))
+    assert {r.observation_id for r in recs} == {"a", "b"}

backend-compliance/tests/test_onboarding_endpoint.py

@@ -73,6 +73,17 @@ def test_partial_signal_surfaces_as_indication_and_is_still_asked():
     assert "secure_development_lifecycle" in asked or "secure_development_lifecycle" in d["capability_delta"]
 
 
+def test_questions_carry_curated_text_and_human_labels():
+    # the curated why_asked from the transition pattern must reach the question (not the generic
+    # fallback "Keine Anhaltspunkte ... klären"), and surfaced capabilities get human labels.
+    body = dict(_BODY, certifications=["ISO27001"], target="TISAX", scanner_findings=[])
+    r = _client.post("/onboarding/advisor-start", json=body)
+    assert r.status_code == 200, r.text
+    d = r.json()
+    assert any("Keine Anhaltspunkte" not in q["why"] for q in d["top_5_questions"])   # real expert text surfaced
+    assert d["capability_labels"].get("vda_isa_self_assessment") == "VDA-ISA-Selbstauskunft"
+
+
 def test_unknown_target_is_404():
     body = dict(_BODY, target="NOPE")
     r = _client.post("/onboarding/advisor-start", json=body)