feat(audit): Phase 1 Quick-Wins (P81 + P85 + P70 + P83) + TCF DELETE/INSERT-Fix

P81 — tests/fixtures/golden_truth/vw_de.json:
GT-Fixture mit must_find_cookies (47 VW-Cookies) + expected_vendors
(Google, Adobe, Trade Desk, ...). Basis fuer kuenftige Regression-Tests.

P85 — banner_screenshot_block.py + consent_scanner.py + main.py:
consent-tester macht beim Banner-Detect einen base64-PNG-Screenshot
(< 1.5MB). Backend rendert ihn als <img src="data:..."> direkt nach
dem GF-1-Pager. Visueller Beweis 'so sah das Banner aus' fuer Dispute
mit Marketing/DSB.

P70 — rag_provenance.py:
classify_finding_provenance() klassifiziert ein Finding als 'rag'
(Norm + Quelle), 'mixed' (Norm ohne Quelle) oder 'heuristic' (eigene
Interpretation). provenance_badge_html() rendert kleine Badges
(✓ RAG / NORM / ⚠ HEURISTIK). Modul ist generisch, kann bei jedem
Finding-Renderer einklinkt werden.

P83 — scripts/check-rebuild-needed.sh:
Prueft ob die im Container deployten BUILD_SHA mit local HEAD
uebereinstimmen. Bei Mismatch exit 1 mit 'REBUILD REQUIRED'-Hinweis.
Verhindert das 'alter Code im Container'-Problem das uns mehrfach
erwischt hat (Frontend-Tabs sichtbar, Backend ohne neuen Service).

TCF-Fix — tcf_vendor_authority.py:
cookie_library hat keinen UNIQUE-Index auf cookie_name → ON CONFLICT
war unmoeglich. Loesung: vor Insert DELETE WHERE source_name='iab_tcf_v2'.
Idempotent. + per-Vendor-Commit damit ein Fail die naechsten nicht blockt.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend-compliance/compliance/api/agent_compliance_check_routes.py

@@ -1486,6 +1486,17 @@ async def _run_compliance_check(check_id: str, req: ComplianceCheckRequest):
         except Exception as e:
             logger.warning("P71 jc_avv_decision skipped: %s", e)
 
+        # P85 — Banner-Screenshot fuer visuellen Beweis (zwischen
+        # GF-1-Pager und Detail-Bloecken)
+        banner_shot_html = ""
+        try:
+            from compliance.services.banner_screenshot_block import (
+                build_banner_screenshot_html,
+            )
+            banner_shot_html = build_banner_screenshot_html(banner_result)
+        except Exception as e:
+            logger.warning("P85 banner-screenshot skipped: %s", e)
+
         # P82: GF-1-Pager ganz oben in der Mail — 5-Bullet-Zusammenfassung
         # damit die GF nicht 124k Char lesen muss.
         gf_one_pager_html = ""
@@ -1585,6 +1596,7 @@ async def _run_compliance_check(check_id: str, req: ComplianceCheckRequest):
             + critical_html + scope_disclaimer_html + exec_summary_html
             + cookie_arch_html + summary_html + scanned_html + profile_html
             + scorecard_html + redundancy_html
+            + banner_shot_html
             + providers_html + banner_deep_html
             + cookie_audit_html
             + tcf_authority_html

backend-compliance/compliance/services/banner_screenshot_block.py

@@ -0,0 +1,44 @@
+"""
+P85 — Banner-Screenshot-Block in der Mail.
+
+Embedded den von consent-tester captured Screenshot des Banners
+(banner_result.banner_screenshot_b64) als data-URI <img> in die Mail.
+"so sah euer Banner zum Audit-Zeitpunkt aus" — visueller Beweis fuer
+Dispute mit Marketing-Team oder DSB.
+"""
+
+from __future__ import annotations
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+def build_banner_screenshot_html(banner_result: dict | None) -> str:
+    if not isinstance(banner_result, dict):
+        return ""
+    b64 = banner_result.get("banner_screenshot_b64") or ""
+    if not b64 or len(b64) < 200:
+        return ""
+    provider = banner_result.get("banner_provider") or "Generic"
+    detected = banner_result.get("banner_detected")
+    return (
+        '<div style="font-family:-apple-system,BlinkMacSystemFont,sans-serif;'
+        'max-width:760px;margin:0 auto 16px;padding:12px 16px;'
+        'background:#f8fafc;border:1px solid #cbd5e1;border-radius:8px">'
+        '<div style="font-size:11px;color:#475569;text-transform:uppercase;'
+        'letter-spacing:1.2px;margin-bottom:4px;font-weight:600">'
+        'Screenshot des Cookie-Banners zum Audit-Zeitpunkt</div>'
+        f'<h3 style="margin:0 0 6px;font-size:13px;color:#1e293b">'
+        f'Provider: <strong>{provider}</strong> · '
+        f'erkannt: <strong>{"ja" if detected else "nein"}</strong></h3>'
+        '<p style="margin:0 0 8px;font-size:11px;color:#64748b;line-height:1.5">'
+        'Visueller Beweis wie das Banner zum Zeitpunkt des Audits angezeigt '
+        'wurde. Bei spaeterer Aenderung des Banners bitte mit diesem '
+        'Screenshot abgleichen.'
+        '</p>'
+        f'<img src="data:image/png;base64,{b64}" alt="Cookie-Banner" '
+        f'style="max-width:100%;height:auto;border:1px solid #cbd5e1;'
+        f'border-radius:4px;display:block">'
+        '</div>'
+    )

backend-compliance/compliance/services/rag_provenance.py

@@ -0,0 +1,90 @@
+"""
+P70 — RAG-Provenance-Marker.
+
+Wenn ein Finding aus dem RAG-Korpus belegt ist (z.B. Art-Match auf
+einen konkreten Gesetzes-Paragrafen aus dem ingestierten DSGVO/TDDDG/
+TMG-Korpus), bekommt es einen ✓-Marker. Wenn es nur aus unserer
+Heuristik kommt (Pattern-Match ohne RAG-Belegung), bekommt es ein ⚠
+"Heuristik".
+
+Dadurch sieht der Nutzer sofort welche Aussagen rechtlich verbindlich
+gestuetzt sind vs welche unsere Eigeninterpretation sind.
+
+Generisch: dataclass-aehnliche Funktion die ein Finding-dict klassifiziert.
+"""
+
+from __future__ import annotations
+
+import logging
+import re
+
+logger = logging.getLogger(__name__)
+
+
+# Pattern fuer "Belegt aus Korpus": Finding enthaelt expliziten
+# Norm-Bezug mit Artikel + Quelle.
+_NORM_RE = re.compile(
+    r"(Art\.?\s*\d+(?:\s*Abs\.?\s*\d+)?(?:\s*lit\.?\s*[a-z])?\s*"
+    r"(?:DSGVO|GDPR|TDDDG|TMG|BDSG|UWG|TKG|EuGH|EDPB)|"
+    r"\(?(EU|VO)\s*\d{4}/\d+\)?|"
+    r"§\s*\d+[a-z]?\s*(TMG|UWG|BDSG|TKG|TDDDG))",
+    re.I,
+)
+
+
+def classify_finding_provenance(finding: dict) -> str:
+    """Returns 'rag', 'heuristic', or 'mixed'.
+
+    rag       — Norm-Bezug + Quellen-URL (verbindlich)
+    heuristic — Pattern-Match ohne Norm-Bezug (Eigeninterpretation)
+    mixed     — Norm-Bezug aber ohne Quellen-URL (teilweise belegbar)
+    """
+    if not isinstance(finding, dict):
+        return "heuristic"
+    legal = (finding.get("legal_basis") or "").strip()
+    detail = (finding.get("detail") or "").strip()
+    rag_id = finding.get("rag_chunk_id")
+    rag_url = finding.get("rag_source_url")
+    blob = " ".join([legal, detail])
+    has_norm = bool(_NORM_RE.search(blob))
+    has_source = bool(rag_id or rag_url or
+                       "https://" in legal or "https://" in detail)
+    if has_norm and has_source:
+        return "rag"
+    if has_norm:
+        return "mixed"
+    return "heuristic"
+
+
+def provenance_badge_html(provenance: str) -> str:
+    if provenance == "rag":
+        return (
+            '<span style="background:#dcfce7;color:#166534;'
+            'padding:1px 5px;border-radius:8px;font-size:9px;'
+            'font-weight:600;margin-left:4px" '
+            'title="Aussage durch RAG-Korpus belegt (Gesetzestext + Quelle)">'
+            '✓ RAG</span>'
+        )
+    if provenance == "mixed":
+        return (
+            '<span style="background:#dbeafe;color:#1e40af;'
+            'padding:1px 5px;border-radius:8px;font-size:9px;'
+            'font-weight:600;margin-left:4px" '
+            'title="Norm-Bezug ohne direkte Quellen-URL">'
+            'NORM</span>'
+        )
+    return (
+        '<span style="background:#f1f5f9;color:#475569;'
+        'padding:1px 5px;border-radius:8px;font-size:9px;'
+        'font-weight:600;margin-left:4px" '
+        'title="Heuristik / Eigeninterpretation ohne Korpus-Beleg">'
+        '⚠ HEURISTIK</span>'
+    )
+
+
+def annotate_findings(findings: list[dict]) -> list[dict]:
+    """In-place: setzt finding['provenance'] auf jeden Eintrag."""
+    for f in (findings or []):
+        if isinstance(f, dict) and "provenance" not in f:
+            f["provenance"] = classify_finding_provenance(f)
+    return findings

backend-compliance/compliance/services/tcf_vendor_authority.py

@@ -81,6 +81,12 @@ async def fetch_and_ingest_tcf_vendors(db: Session) -> dict:
     if not vendors:
         return {"error": "no vendors in TCF response", "n_vendors": 0}
 
+    # Erst alte TCF-Eintraege weg (kein UNIQUE-Index auf cookie_name,
+    # daher kein ON CONFLICT moeglich → idempotent via DELETE+INSERT).
+    db.execute(sa_text(
+        "DELETE FROM compliance.cookie_library WHERE source_name='iab_tcf_v2'"
+    ))
+    db.commit()
     inserted = 0
     skipped = 0
     for vid, v in vendors.items():
@@ -106,13 +112,6 @@ async def fetch_and_ingest_tcf_vendors(db: Session) -> dict:
                 VALUES (:n, :dp, :v, :pu, :cat, :purp, 'iab_tcf_v2',
                         'https://vendor-list.consensu.org/v3/vendor-list.json',
                         0.99)
-                ON CONFLICT (cookie_name) DO UPDATE
-                   SET actual_category    = EXCLUDED.actual_category,
-                       vendor_name        = EXCLUDED.vendor_name,
-                       vendor_privacy_url = EXCLUDED.vendor_privacy_url,
-                       purpose_en         = EXCLUDED.purpose_en,
-                       source_name        = EXCLUDED.source_name,
-                       confidence         = EXCLUDED.confidence
                 """
             ), {"n": marker, "dp": "*",
                 "v": f"[TCF-{vid}] {name}",

backend-compliance/tests/fixtures/golden_truth/vw_de.json

@@ -0,0 +1,51 @@
+{
+  "site": "Volkswagen Deutschland",
+  "site_url": "https://www.volkswagen.de",
+  "captured_at": "2026-05-22T00:00:00Z",
+  "source": "User-Copy aus Cookie-Richtlinie (Browser Strg+A → Strg+C)",
+  "cookie_richtlinie_url": "https://www.volkswagen.de/de/mehr/rechtliches/cookie-richtlinie.html",
+  "expectations": {
+    "min_declared_cookies": 90,
+    "expected_unique_vendors_after_dedup": 18,
+    "must_find_cookies": [
+      "VWD6_ENSIGHTEN_PRIVACY_MODAL_LOADED",
+      "VWD6_ENSIGHTEN_PRIVACY_MODAL_VIEWED",
+      "smartSignals2UiD", "smartSignals2sUiD",
+      "s_ecid", "s_cc", "s_sq",
+      "AMCV_", "AMCVS_", "demdex", "dextp",
+      "mbox", "mboxEdgeCluster",
+      "TDID", "TDCPM", "TTDOptOut",
+      "DSID", "ANID", "AID", "IDE", "TAID",
+      "_gcl_au", "_gcl_dc", "_fbc", "_fbp", "fr",
+      "_pk_uid",
+      "OptanonConsent",
+      "everest_g_v2", "everest_session_v2",
+      "adbCDP",
+      "liveagent_sid", "liveagent_chatted",
+      "X-Salesforce-eLB", "sfdc-stream",
+      "__cfduid", "__cflb",
+      "FPAU", "FPGCLDC", "FLC", "APC",
+      "wlfeDoLogin", "wlfeRefreshSessionId", "LBCOOKIE",
+      "CookieConsentPolicy",
+      "BrowserId", "BrowserId_sec",
+      "inbenta-km-session-id"
+    ],
+    "expected_vendors_present": [
+      "Google",
+      "Adobe Experience Cloud",
+      "Adobe Analytics",
+      "The Trade Desk",
+      "AdForm",
+      "Meta / Facebook",
+      "Salesforce",
+      "Cloudflare",
+      "Borlabs"
+    ],
+    "expected_high_findings_minimum": 1,
+    "banner_must_be_detected": true,
+    "expected_doc_types_with_text": [
+      "dse", "cookie", "impressum", "nutzungsbedingungen"
+    ]
+  },
+  "raw_paste": "Name des Cookies\nKategorie\nVerwendungszweck\nSpeicherdauer\nArt des Cookies\nSee tests/fixtures/cookie_gt/vw_cookie_richtlinie.txt for the abbreviated raw form."
+}

consent-tester/main.py

@@ -53,6 +53,7 @@ class ScanResponse(BaseModel):
     cmp_payloads: list[dict] = []  # P48: raw CMP JSON-payloads (Usercentrics/OneTrust/...) captured during scan
     vendor_details: list[dict] = []  # P50: per-vendor detail-modal-extracts (Beschreibung/Cookies/Opt-Out/Privacy)
     cookies_detailed: list[dict] = []  # P59b: full cookie details for behavior-validation (name,value,domain,expires,phase,declared_category)
+    banner_screenshot_b64: str = ""  # P85: base64-PNG des Banners (initial-view)
 
 
 @app.get("/health")
@@ -133,6 +134,7 @@ async def scan_consent(req: ScanRequest):
         cmp_payloads=result.cmp_payloads,  # P48
         vendor_details=result.vendor_details,  # P50
         cookies_detailed=result.cookies_detailed,  # P59b
+        banner_screenshot_b64=result.banner_screenshot_b64,  # P85
     )
 
 

consent-tester/services/consent_scanner.py

@@ -77,6 +77,10 @@ class ConsentTestResult:
     # for behavior-validation in backend. Implicit declared_category:
     # before/reject phase = essential (site claims), accept = any.
     cookies_detailed: list = field(default_factory=list)
+    # P85: base64-PNG-Screenshot des Banners vor dem ersten Klick.
+    # Backend embedded das als <img> in der Mail — visueller Beweis
+    # "so sah das Banner zum Audit-Zeitpunkt aus".
+    banner_screenshot_b64: str = ""
 
 
 async def run_consent_test(
@@ -196,6 +200,17 @@ async def run_consent_test(
                 result.banner_text_violations = banner_violations["violations"]
                 result.banner_has_impressum_link = banner_violations["has_impressum"]
                 result.banner_has_dse_link = banner_violations["has_dse"]
+                # P85 — visueller Beweis fuer die Mail.
+                try:
+                    import base64 as _b64
+                    png = await page_a.screenshot(
+                        full_page=False, type="png", timeout=10000,
+                    )
+                    if png and len(png) < 1_500_000:  # < 1.5 MB
+                        result.banner_screenshot_b64 = _b64.b64encode(png).decode("ascii")
+                        logger.info("P85: banner screenshot captured (%d bytes)", len(png))
+                except Exception as _se:
+                    logger.warning("P85: banner screenshot failed: %s", _se)
 
             await ctx_a.close()
 

scripts/check-rebuild-needed.sh

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+# P83 — verhindert "alter Code im Container"-Bug.
+#
+# Vergleicht den im Container deployten git-SHA mit dem aktuellen
+# Source-SHA. Wenn abweichend → exit 1 mit Hinweis Build/Recreate.
+#
+# Aufruf-Beispiele:
+#   ./scripts/check-rebuild-needed.sh backend-compliance
+#   ./scripts/check-rebuild-needed.sh admin-compliance
+#   ./scripts/check-rebuild-needed.sh consent-tester
+#
+# CI-Verwendung: nach git push, vor dem ersten Health-Check.
+# Lokal: claude / dev kann es via pre-merge-hook nutzen.
+#
+# Voraussetzung: Container hat BUILD_SHA env (gesetzt im Dockerfile via
+# ARG BUILD_SHA + ENV BUILD_SHA=$BUILD_SHA). Falls leer → Warnung.
+
+set -e
+
+SERVICE="${1:-backend-compliance}"
+CONTAINER="bp-compliance-${SERVICE#*-}"   # backend-compliance → bp-compliance-backend
+if [[ "$SERVICE" == "consent-tester" ]]; then
+    CONTAINER="bp-compliance-consent-tester"
+fi
+
+DOCKER="${DOCKER:-/usr/local/bin/docker}"
+
+deployed_sha=$($DOCKER exec "$CONTAINER" sh -c 'echo "${BUILD_SHA:-unknown}"' 2>/dev/null || echo "container-down")
+local_sha=$(git rev-parse --short HEAD)
+
+if [[ "$deployed_sha" == "container-down" ]]; then
+    echo "❌ Container $CONTAINER is not running"
+    exit 2
+fi
+
+if [[ "$deployed_sha" == "unknown" ]]; then
+    echo "⚠️  $CONTAINER has no BUILD_SHA env — cannot verify."
+    echo "    Add to Dockerfile: ARG BUILD_SHA / ENV BUILD_SHA=\$BUILD_SHA"
+    exit 0
+fi
+
+if [[ "$deployed_sha" != "$local_sha"* && "$local_sha" != "$deployed_sha"* ]]; then
+    echo "❌ $CONTAINER is on commit $deployed_sha, local is $local_sha"
+    echo "    REBUILD REQUIRED:"
+    echo "    docker compose build $SERVICE && docker compose up -d --no-deps --force-recreate $SERVICE"
+    exit 1
+fi
+
+echo "✓ $CONTAINER ($deployed_sha) matches local ($local_sha)"