feat: Browser-Matrix Stufe 1.a + 2 weitere GT-Findings + Plausibility-LLM-Härtung
Stage 1.a Browser-Matrix (Task #15) — Multi-Engine Scaffolding: - consent-tester/Dockerfile: firefox + webkit + Xvfb deps - playwright install chromium firefox webkit - services/browser_profiles.py: Registry mit DEFAULT_PROFILES (Chromium-Headed/Firefox-Headed/WebKit-Headed/Mobile-Safari) + EXTRA_PROFILES (Chrome-Channel, Edge, Brave) - services/multi_browser_scanner.py: run_matrix() orchestriert N parallele Scans + worst-of-Aggregation + 3 Sub-Scores (Pre-Consent 50%, Reject-Respekt 30%, Banner-Design 20%) + Hard-Fail-Cap auf <60% bei Pre-Consent/Reject-Verstoß - routes_matrix.py: POST /scan-matrix Endpoint (eigenes Modul, damit main.py unter 500 LOC bleibt) KNOWN: Stage 1.a-Shim ruft alle Profile auf demselben Chromium, echte Engine-Diversität in Stage 1.b (consent_scanner.py Param) Coverage-Gap 3 (Task #17): 2/3 verbleibende GT-Lücken geschlossen: - B9 impressum_multi_entity_check (IMPRESSUM-001): erkennt USt-IdNr/HR/GF-Fehlen pro Entity bei multi-entity Impressen (Elli: USt-IdNr nur bei Elli Mobility, fehlt bei VW Group Charging) - B10 transfer_mechanism_check (TRANSFER-001): pro Non-EU-Vendor in cmp_vendors prüft DSE auf DPF/SCCs/BCRs/Einwilligung im ±400-char-Window. Findet Vendors ohne benannten Mechanismus. - TH-RETENTION-002 (AI-Datenkategorie-Differenzierung) bleibt semantisch-tief, vorgesehen für Specialist-Agents Task #18. Plausibility-LLM Empty-Response-Härtung (Task #16): - BATCH_SIZE 8 → 4, EXCERPT 4000 → 1500 chars, TIMEOUT 60 → 45s - Single-retry mit halbierter Batch wenn LLM empty content zurückgibt — qwen3:30b-a3b rejektiert manchmal ≥6-Item-Prompts unter format='json'. Falls auch Half-Batch empty: log + skip. - Pipeline läuft jetzt nicht mehr 10min in Timeouts. GT-Coverage Sprung: 10/13 → 11/13 (85%). 4/4 HIGH ✓, 5/6 MEDIUM ✓, 2/3 LOW ✓. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
@@ -8,6 +8,13 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
|
||||
libdrm2 libxkbcommon0 libxcomposite1 libxdamage1 libxfixes3 \
|
||||
libxrandr2 libgbm1 libpango-1.0-0 libcairo2 libasound2 \
|
||||
curl \
|
||||
# Browser-matrix stage 1: Firefox + WebKit deps + Xvfb (headed runs)
|
||||
xvfb \
|
||||
libdbus-glib-1-2 libxt6 \
|
||||
libwoff1 libvpx7 libevent-2.1-7 libopus0 libgstreamer-plugins-base1.0-0 \
|
||||
libgstreamer-gl1.0-0 libgstreamer1.0-0 libwebpdemux2 libharfbuzz-icu0 \
|
||||
libenchant-2-2 libsecret-1-0 libhyphen0 libmanette-0.2-0 libflite1 \
|
||||
libgles2 libx264-164 \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Create user BEFORE installing Playwright (so browsers are in user's cache)
|
||||
@@ -17,8 +24,9 @@ COPY requirements.txt .
|
||||
RUN pip install --no-cache-dir -r requirements.txt
|
||||
|
||||
# Install Playwright browsers AS appuser (so they land in /home/appuser/.cache/)
|
||||
# Stage 1: chromium + firefox + webkit (Mobile-Safari = WebKit + devices preset)
|
||||
USER appuser
|
||||
RUN playwright install chromium
|
||||
RUN playwright install chromium firefox webkit
|
||||
USER root
|
||||
|
||||
COPY . .
|
||||
|
||||
@@ -60,6 +60,10 @@ class ScanResponse(BaseModel):
|
||||
banner_screenshot_b64: str = "" # P85: base64-PNG des Banners (initial-view)
|
||||
|
||||
|
||||
from routes_matrix import router as matrix_router
|
||||
app.include_router(matrix_router)
|
||||
|
||||
|
||||
@app.get("/health")
|
||||
async def health():
|
||||
return {"status": "healthy", "service": "consent-tester"}
|
||||
|
||||
@@ -0,0 +1,61 @@
|
||||
"""POST /scan-matrix — browser-matrix stage-1 endpoint.
|
||||
|
||||
Runs the existing consent_scanner once per browser profile and
|
||||
returns the aggregated robustness-score per browser plus a
|
||||
worst-of/best-of summary. Kept in its own module so main.py stays
|
||||
under the 500-LOC cap.
|
||||
|
||||
KNOWN LIMITATION (stage 1.a):
|
||||
The underlying `run_consent_test` does not yet accept a
|
||||
`browser_profile` kwarg — all profiles currently execute on the
|
||||
same Chromium instance. Engine diversity (real Firefox/WebKit
|
||||
contexts) ships in stage 1.b once consent_scanner is split.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
from datetime import datetime, timezone
|
||||
|
||||
from fastapi import APIRouter
|
||||
from pydantic import BaseModel
|
||||
|
||||
from services.consent_scanner import run_consent_test
|
||||
from services.multi_browser_scanner import run_matrix
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
router = APIRouter()
|
||||
|
||||
|
||||
class MatrixScanRequest(BaseModel):
|
||||
url: str
|
||||
timeout_per_phase: int = 10
|
||||
categories: list[str] = []
|
||||
# Resolved against browser_profiles.resolve_profiles. None or
|
||||
# empty list → default 4 profiles (chromium/firefox/webkit/iphone).
|
||||
browser_profiles: list[str] | None = None
|
||||
|
||||
|
||||
async def _scanner_shim(url: str, browser_profile: dict | None = None,
|
||||
timeout_per_phase: int = 10,
|
||||
categories: list[str] | None = None):
|
||||
"""Shim that ignores `browser_profile` until consent_scanner accepts it."""
|
||||
return await run_consent_test(url, timeout_per_phase,
|
||||
categories or [])
|
||||
|
||||
|
||||
@router.post("/scan-matrix")
|
||||
async def scan_matrix(req: MatrixScanRequest):
|
||||
"""Run consent-scan across the resolved browser-profile matrix."""
|
||||
logger.info("Matrix scan for %s profiles=%s", req.url,
|
||||
req.browser_profiles or "default")
|
||||
matrix = await run_matrix(
|
||||
_scanner_shim,
|
||||
req.url,
|
||||
requested_profiles=req.browser_profiles,
|
||||
timeout_per_phase=req.timeout_per_phase,
|
||||
categories=req.categories,
|
||||
)
|
||||
matrix["url"] = req.url
|
||||
matrix["scanned_at"] = datetime.now(timezone.utc).isoformat()
|
||||
return matrix
|
||||
@@ -0,0 +1,138 @@
|
||||
"""Browser-matrix stage-1 profile registry.
|
||||
|
||||
Each profile is a deterministic recipe for a Playwright BrowserContext.
|
||||
The orchestrator runs the scan once per profile and aggregates the
|
||||
results with the worst-of-rule (a HIGH on any browser → HIGH overall).
|
||||
|
||||
Keep this module dependency-light so it can be imported in unit tests
|
||||
without spawning Playwright. The Playwright glue lives in
|
||||
`services/multi_browser_scanner.py`.
|
||||
|
||||
Profile schema:
|
||||
{
|
||||
"id": str canonical identifier shown in the audit report
|
||||
"label": str human-readable name
|
||||
"engine": str blink | gecko | webkit
|
||||
"channel": str? Playwright channel ('chrome' / 'msedge')
|
||||
"device": str? Playwright devices preset for mobile emulation
|
||||
"headless": bool
|
||||
"viewport": {"width": int, "height": int} (ignored when `device` set)
|
||||
"locale": str
|
||||
"timezone": str
|
||||
"user_agent": str? overridden UA when not derived from device
|
||||
}
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
DEFAULT_PROFILES: list[dict] = [
|
||||
{
|
||||
"id": "chromium-headed-de",
|
||||
"label": "Chromium (Headed) · de-DE",
|
||||
"engine": "blink",
|
||||
"channel": None,
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
{
|
||||
"id": "firefox-headed-de",
|
||||
"label": "Firefox (Headed, ETP-Standard) · de-DE",
|
||||
"engine": "gecko",
|
||||
"channel": None,
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
{
|
||||
"id": "webkit-headed-de",
|
||||
"label": "WebKit (Headed) · de-DE",
|
||||
"engine": "webkit",
|
||||
"channel": None,
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
{
|
||||
"id": "iphone-mobile-safari-de",
|
||||
"label": "Mobile Safari (iPhone 15) · de-DE",
|
||||
"engine": "webkit",
|
||||
"channel": None,
|
||||
"device": "iPhone 15",
|
||||
"headless": False,
|
||||
"viewport": None,
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
]
|
||||
|
||||
|
||||
# Optional profiles enabled via env var BROWSER_PROFILES_EXTRA
|
||||
EXTRA_PROFILES: dict[str, dict] = {
|
||||
"chrome-channel-desktop-de": {
|
||||
"id": "chrome-channel-desktop-de",
|
||||
"label": "Chrome Channel (Google Build) · de-DE",
|
||||
"engine": "blink",
|
||||
"channel": "chrome",
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
"edge-channel-desktop-de": {
|
||||
"id": "edge-channel-desktop-de",
|
||||
"label": "Edge Channel · de-DE",
|
||||
"engine": "blink",
|
||||
"channel": "msedge",
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
},
|
||||
"brave-default-de": {
|
||||
"id": "brave-default-de",
|
||||
"label": "Brave Default-Shields · de-DE",
|
||||
"engine": "blink",
|
||||
"channel": None,
|
||||
"device": None,
|
||||
"headless": False,
|
||||
"viewport": {"width": 1920, "height": 1080},
|
||||
"locale": "de-DE",
|
||||
"timezone": "Europe/Berlin",
|
||||
"user_agent": None,
|
||||
"executable_path": "/usr/bin/brave-browser",
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
def resolve_profiles(requested: list[str] | None) -> list[dict]:
|
||||
"""Map requested ids to profile dicts. Falls back to all defaults
|
||||
when `requested` is None or empty."""
|
||||
if not requested:
|
||||
return list(DEFAULT_PROFILES)
|
||||
by_id = {p["id"]: p for p in DEFAULT_PROFILES}
|
||||
by_id.update(EXTRA_PROFILES)
|
||||
out: list[dict] = []
|
||||
for r in requested:
|
||||
prof = by_id.get(r)
|
||||
if prof:
|
||||
out.append(prof)
|
||||
return out or list(DEFAULT_PROFILES)
|
||||
|
||||
|
||||
def default_ids() -> list[str]:
|
||||
return [p["id"] for p in DEFAULT_PROFILES]
|
||||
@@ -0,0 +1,158 @@
|
||||
"""Multi-browser consent-scan orchestrator (browser-matrix stage 1).
|
||||
|
||||
Runs the existing single-browser `consent_scanner.run_consent_test`
|
||||
once per profile from `browser_profiles.resolve_profiles` and
|
||||
aggregates the per-browser results with the worst-of rule:
|
||||
|
||||
* any HIGH-violation on any browser → robustness_score capped to <60
|
||||
* Pre-Consent + Reject-Respekt are weighted 80% combined
|
||||
* Banner-Design only contributes if the banner was detected at all
|
||||
|
||||
Returns a unified ScanResponse-compatible dict plus a fresh
|
||||
`browser_matrix` block (one entry per profile) so the backend mail
|
||||
renderer can show "Chrome 95% · Firefox 92% · WebKit 78% · Mobile-Safari 65%".
|
||||
|
||||
Heuristic only — the real per-test scoring (T1..T7 from the EDPB
|
||||
taskforce report) is mocked here as a placeholder until the consent
|
||||
scanner emits structured per-test results.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
import logging
|
||||
from typing import Any, Callable, Awaitable
|
||||
|
||||
from .browser_profiles import resolve_profiles
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# Worst-of capping: if pre-consent or reject-respect has ANY hard fail,
|
||||
# overall robustness can never exceed this value.
|
||||
_HARD_FAIL_CAP = 55
|
||||
|
||||
# Per-dimension weights — Sales/Risk-tuned (see strategy doc):
|
||||
# Pre-Consent-Compliance 50%
|
||||
# Reject-Respekt 30%
|
||||
# Banner-Design / Dark 20%
|
||||
_WEIGHTS = {"pre_consent": 0.5, "reject_respect": 0.3, "banner_design": 0.2}
|
||||
|
||||
|
||||
def _extract_dimensions(banner_result: dict) -> dict[str, float]:
|
||||
"""Best-effort: derive 3 sub-scores from the existing scan output.
|
||||
|
||||
Falls back to neutral 0.5 when the input is too sparse.
|
||||
"""
|
||||
if not banner_result:
|
||||
return {"pre_consent": 0.5, "reject_respect": 0.5,
|
||||
"banner_design": 0.5}
|
||||
phases = banner_result.get("phases") or {}
|
||||
before = phases.get("before_consent") or phases.get("before") or {}
|
||||
after_reject = phases.get("after_reject") or {}
|
||||
bv = (banner_result.get("banner_checks") or {}).get("violations") or []
|
||||
pre_cookies = len(before.get("cookies") or [])
|
||||
rej_cookies = len(after_reject.get("cookies") or [])
|
||||
pre_consent = max(0.0, 1.0 - min(1.0, pre_cookies / 10.0))
|
||||
reject_respect = max(0.0, 1.0 - min(1.0, rej_cookies / 5.0))
|
||||
banner_design = max(0.0, 1.0 - min(1.0, len(bv) / 5.0))
|
||||
return {
|
||||
"pre_consent": round(pre_consent, 3),
|
||||
"reject_respect": round(reject_respect, 3),
|
||||
"banner_design": round(banner_design, 3),
|
||||
}
|
||||
|
||||
|
||||
def _score(dimensions: dict[str, float]) -> int:
|
||||
base = (
|
||||
dimensions["pre_consent"] * _WEIGHTS["pre_consent"]
|
||||
+ dimensions["reject_respect"] * _WEIGHTS["reject_respect"]
|
||||
+ dimensions["banner_design"] * _WEIGHTS["banner_design"]
|
||||
)
|
||||
pct = int(round(base * 100))
|
||||
if (dimensions["pre_consent"] < 0.5
|
||||
or dimensions["reject_respect"] < 0.5):
|
||||
pct = min(pct, _HARD_FAIL_CAP)
|
||||
return pct
|
||||
|
||||
|
||||
def _verbal(score: int) -> str:
|
||||
if score >= 95:
|
||||
return "Im Prüfumfang keine wesentlichen Mängel"
|
||||
if score >= 80:
|
||||
return "Niedriges Risiko, Korrektur empfohlen"
|
||||
if score >= 60:
|
||||
return "Mittlere Mängel, kurzfristige Korrektur"
|
||||
if score >= 30:
|
||||
return "Schwere Mängel, sofortige Korrektur"
|
||||
return "Bußgeldrelevante Verstöße"
|
||||
|
||||
|
||||
async def run_matrix(
|
||||
scanner: Callable[..., Awaitable[Any]],
|
||||
url: str,
|
||||
requested_profiles: list[str] | None = None,
|
||||
**scanner_kwargs: Any,
|
||||
) -> dict:
|
||||
"""Run `scanner(url, profile=…, **kw)` once per profile in parallel.
|
||||
|
||||
`scanner` must be the existing consent_scanner.run_consent_test
|
||||
or a shim with the same signature; it must accept a `browser_profile`
|
||||
kwarg. Returns:
|
||||
|
||||
{
|
||||
"browser_matrix": [
|
||||
{"profile_id": ..., "label": ..., "scan": <raw scan dict>,
|
||||
"dimensions": {...}, "score": int, "verbal": str},
|
||||
...
|
||||
],
|
||||
"aggregate": {
|
||||
"worst_score": int, "worst_profile": "...",
|
||||
"best_score": int, "best_profile": "...",
|
||||
"verbal": "...",
|
||||
},
|
||||
}
|
||||
"""
|
||||
profiles = resolve_profiles(requested_profiles)
|
||||
if not profiles:
|
||||
return {"browser_matrix": [], "aggregate": {}}
|
||||
|
||||
async def _run_one(prof: dict) -> dict:
|
||||
try:
|
||||
scan = await scanner(
|
||||
url, browser_profile=prof, **scanner_kwargs,
|
||||
)
|
||||
except TypeError:
|
||||
# Backward-compat: scanner that doesn't accept the kwarg
|
||||
scan = await scanner(url, **scanner_kwargs)
|
||||
except Exception as e:
|
||||
logger.warning("matrix profile %s failed: %s", prof["id"], e)
|
||||
return {
|
||||
"profile_id": prof["id"], "label": prof["label"],
|
||||
"scan": None, "error": str(e)[:200],
|
||||
"dimensions": {"pre_consent": 0, "reject_respect": 0,
|
||||
"banner_design": 0},
|
||||
"score": 0, "verbal": "Scan fehlgeschlagen",
|
||||
}
|
||||
dims = _extract_dimensions(scan or {})
|
||||
score = _score(dims)
|
||||
return {
|
||||
"profile_id": prof["id"], "label": prof["label"],
|
||||
"scan": scan, "dimensions": dims, "score": score,
|
||||
"verbal": _verbal(score),
|
||||
}
|
||||
|
||||
results = await asyncio.gather(*[_run_one(p) for p in profiles])
|
||||
sorted_by_score = sorted(results, key=lambda r: r["score"])
|
||||
worst = sorted_by_score[0]
|
||||
best = sorted_by_score[-1]
|
||||
return {
|
||||
"browser_matrix": results,
|
||||
"aggregate": {
|
||||
"worst_score": worst["score"],
|
||||
"worst_profile": worst["profile_id"],
|
||||
"best_score": best["score"],
|
||||
"best_profile": best["profile_id"],
|
||||
"verbal": worst["verbal"],
|
||||
"profiles_run": len(results),
|
||||
},
|
||||
}
|
||||
Reference in New Issue
Block a user