French & European 4G/5G Mobile Proxies: Why Local Telecom Carriers Outperform Global VPNs for Localized SERP & E-Commerce

Modern anti-bot engines, fraud-prevention suites, and localized content delivery networks no longer evaluate incoming HTTP requests purely on static heuristics such as User-Agent strings or TLS ciphers. The modern web security perimeter operates on network-layer intelligence. Platforms such as DataDome (heavily deployed across French e-commerce), Cloudflare Turnstile, Sift, Riskified, and Google Search localization engines analyze the incoming request at Layers 3, 4, and 7 simultaneously.

When web scrapers, competitive intelligence pipelines, and multi-account automation engines route traffic through commercial VPNs or datacenter IP addresses (e.g., OVH, Hetzner, AWS, DigitalOcean), they fail foundational network reputation audits. These requests are intercepted, challenged with hard CAPTCHAs, or fed poisoned, non-localized data.

To achieve persistent, unblockable throughput across French and wider European targets—such as Leboncoin, Amazon.fr, Cdiscount, Fnac, Vinted, and local Google SERPs (Paris, Lyon, Marseille)—infrastructure teams must route through legitimate local telecom carrier space.

This guide provides an exhaustive engineering breakdown of why physical, industrial French 4G/5G hardware on tier-1 mobile networks (Orange, SFR, Free Mobile, Bouygues Telecom) mathematically and architectural outperforms datacenter proxies, commercial VPN networks, and synthetic residential proxy pools.

1. Network Topologies: The Cellular Gateway vs. The VPN Exit Node

To understand why commercial VPNs fail against modern French e-commerce defenses, you must dissect the Layer 3 and Layer 4 differences between a standard VPN server hosted in a datacenter and a true cellular endpoint operating behind a mobile network operator's (MNO) Core Network.

DATACENTER / VPN TOPOLOGY (INSTANTLY FLAGGED):
[ Scraper / Client ] 
       │ (Encrypted WireGuard/OpenVPN Tunnel)
       ▼
[ Datacenter Server (OVH / Hetzner / AWS) ] ── BGP AS16276 / AS24940
       │ 
       ├─ Static IPv4 Range (/24 to /16)
       ├─ TCP MSS: 1460 (Ethernet Standard)
       ├─ No Carrier-Grade NAT (1 IP : 1 Client)
       ▼
[ Target: Leboncoin / DataDome ] ──> ACTION: TCP Reset / 403 Forbidden / Proof-of-Work

REAL 4G/5G CELLULAR TOPOLOGY (TRUSTED):
[ Scraper / Client ]
       │ (SOCKS5 / HTTP Proxy Auth)
       ▼
[ Industrial Modem: Teltonika RUTX50 (Proxym) ]
       │ (5G-NR Sub-6GHz Radio Interface)
       ▼
[ Base Station (eNodeB / gNodeB) ]
       │ (GTP-U Tunnel over Cellular Backhaul)
       ▼
[ French MNO Core: Orange (AS3215) / Free (AS12322) / SFR (AS15557) ]
       │
       ├─ Serving Gateway (SGW) / User Plane Function (UPF)
       ├─ Carrier-Grade NAT (CGNAT) ── 1 Public IP : Thousands of Real Smartphones
       ├─ TCP MSS: Clamped to 1360–1400 (Cellular Overhead)
       ├─ Dynamic IP Churn via Radio Resource Reallocation
       ▼
[ Target: Leboncoin / DataDome ] ──> ACTION: 200 OK (Zero Friction)

The Inherent Flaws of Datacenter VPNs

Datacenter VPN architectures suffer from structural markers that trigger immediate blacklisting:

  • BGP Autonomous System Number (ASN) Classification: Every IP block announced to the global BGP routing table belongs to an Autonomous System. ASNs owned by hosting providers—such as OVH (AS16276), Hetzner (AS24940), Scaleway (AS12876), or DigitalOcean (AS14061)—are categorized as "Hosting/Datacenter" by IP intelligence registries (IPinfo, MaxMind, DB-IP). Security software queries these registries in real time. If a visitor to leboncoin.fr arrives via an OVH ASN, the session is assigned an immediate high-risk score because consumer users never browse from a datacenter.
  • Dedicated Allocation Footprint (1:1 IP-to-User Ratio): Datacenter VPN nodes typically assign a static public IP to a user, or route multiple VPN subscribers through a single public IP. Because the traffic contains disparate browser fingerprints across contradictory operating systems, machine-learning models isolate these endpoints as automated proxy gateways.
  • Predictable Subnet Ranges: Datacenter IPs are sold in contiguous CIDR blocks (e.g., /24 or /22). Anti-bot systems block or rate-limit the entire subnet the moment malicious automation is detected on a single IP within that block.

The Physics of Carrier-Grade NAT (CGNAT)

French MNOs (Orange, SFR, Free, Bouygues) mitigate IPv4 address exhaustion using Carrier-Grade NAT (RFC 6598, 100.64.0.0/10 shared address space).

In an MNO network:

  1. The cellular modem (e.g., an industrial Teltonika RUTX50) connects over 3GPP protocols to an eNodeB (4G LTE) or gNodeB (5G NR) tower.
  2. The user equipment is assigned an ephemeral private IP address from the carrier's internal CGNAT pool.
  3. Traffic is encapsulated inside a GTP-U (GPRS Tunnelling Protocol User Plane) tunnel and transmitted back to the carrier's Packet Data Network Gateway (PGW) in 4G Evolved Packet Core (EPC) or the User Plane Function (UPF) in 5G Core (5GC).
  4. The carrier's NAT pools translate the internal GTP packets to a shared public IP address.

Under this architecture, a single public IPv4 address on Orange (AS3215) is shared simultaneously by thousands of active, legitimate mobile devices—ranging from iPhones browsing Instagram to Android phones purchasing tickets on SNCF Connect.

Anti-bot networks cannot blanket-ban a mobile carrier's public CGNAT IP address. If DataDome, Cloudflare, or Akamai were to ban an active Orange or Free Mobile CGNAT IP, they would simultaneously drop hundreds or thousands of high-value human shoppers, causing immediate revenue loss for their enterprise e-commerce clients.

2. The Walled Gardens of French E-Commerce

French digital platforms enforce strict geolocation and anti-automation filters across Europe. Their security postures are calibrated to domestic fraud trends, regional inventory limits, and consumer protection regulations.

+--------------------+-------------------------+-----------------------------------------+
| Platform           | Primary Defense Engine  | Primary Blocking Heuristic              |
+--------------------+-------------------------+-----------------------------------------+
| Leboncoin          | DataDome + Custom ML    | ASN type, Device Fingerprint, TCP MSS   |
| Vinted France      | Cloudflare Enterprise   | JA4 Fingerprinting, IP Reputation, UULE |
| Amazon.fr          | Internal AWS Anti-Bot   | Geo-Radius Latency, Account Clustering  |
| Fnac / Darty       | PerimeterX / HUMAN      | Behavioral Biometrics, Sensor Data      |
| Cdiscount          | Shape / F5 Distributed  | L7 TLS Signatures, ASN Distance Models  |
+--------------------+-------------------------+-----------------------------------------+

Leboncoin (leboncoin.fr)

Leboncoin is the highest-traffic classifieds marketplace in France and one of DataDome's premier showcase deployments. DataDome inspects client connections through an edge-integrated API module that evaluates:

  • TCP/IP Handshake Properties: SYN packet parameters (TTL, Window Size, IP Options).
  • ASN Integrity: Immediate challenge/drop for any non-residential, non-mobile ASN. Foreign residential IPs (e.g., a Comcast or Deutsche Telekom IP requesting French real-estate listings) receive elevated suspicion scores.
  • Client-Side Signals: Dynamic JS payloads measuring device orientation, battery API, canvas hashes, and mouse dynamics.

If you attempt to scrape Leboncoin listing pages over an AWS or Hetzner VPN, DataDome issues a 403 Forbidden payload with an embedded tracking hash within 1 to 5 requests. Over a native French mobile carrier IP (such as Orange or Bouygues), DataDome's risk engine defaults to an allow-list posture because the request profile aligns with the predominant human profile on French mobile networks.

Amazon France (amazon.fr)

Amazon applies complex fraud modeling to search, scraping, and purchase actions. Amazon.fr restricts deals, lightning promotions, and specific delivery estimation calculators based on localized postal codes (Code Postal).

When a client queries Amazon.fr from a foreign VPN (even a high-speed VPN located in Frankfurt or London):

  • The delivery calculator alters buy-box eligibility, frequently marking domestic French goods as unavailable or adding international shipping flags.
  • Aggressive pagination through search queries triggers automated CAPTCHAs (Type the characters you see in this image) after 15–30 requests.
  • When routed through a physical French 5G interface with an IP belonging to Free Mobile (AS12322) or SFR (AS15557), search scraping pipelines can cycle through thousands of SERP pages without encountering CAPTCHAs, because the system treats the traffic as a high-density mobile user navigating on a cellular data plan.

Fnac, Darty, and Cdiscount

These retail engines leverage behavioral analytics from HUMAN Security (formerly PerimeterX) and F5 Distributed Cloud (Shape Security). These platforms track the latency differential between the client's DNS resolution location and the public IP address:

$$\Delta_{\text{geo}} = \text{GeoDistance}(\text{IP}_{\text{Client}}, \text{IP}_{\text{DNSResolv}})$$

If a proxy user routes traffic through a French IP proxy server while their DNS requests resolve via Cloudflare (1.1.1.1) in Ashburn, Virginia, this cross-layer mismatch triggers a bot challenge. Physical cellular proxies terminated on localized French routers bypass this by querying the carrier's upstream DNS infrastructure (e.g., Orange DNS servers at 80.10.246.2 and 80.10.246.129).

3. Deep-Dive into French Telecom ASNs

Deploying automation across France requires understanding the operational topology of the country's primary cellular providers. Each operator maintains distinct IP routing architectures, CGNAT allocations, and core network configurations.

FRENCH OPERATOR BGP & CGNAT ARCHITECTURE:

  [ Orange S.A. ]           [ SFR S.A. ]       [ Free Mobile ]      [ Bouygues ]
     AS3215                   AS15557              AS12322             AS5410
        │                        │                    │                  │
 ┌──────┴──────┐          ┌──────┴──────┐      ┌──────┴──────┐    ┌──────┴──────┐
 │ NAT444      │          │ Mixed IPv4/ │      │ 464XLAT /   │    │ Carrier NAT │
 │ Massive     │          │ CGNAT Pool  │      │ NAT64 Core  │    │ Enterprise  │
 │ Enterprise  │          │ Dynamic     │      │ Aggressive  │    │ APN Support │
 │ Subnets     │          │ Churn       │      │ Sharing     │    │             │
 └─────────────┘          └─────────────┘      └─────────────┘    └─────────────┘

Orange S.A. (AS3215)

Orange is the incumbent telecommunications provider in France, retaining the largest contiguous blocks of both fixed and mobile IP allocations in Western Europe.

  • Routing Architecture: Orange operates AS3215. While AS3215 houses both FTTH (fiber) and mobile endpoints, their mobile gateway systems map to specific regional peering points (primarily Paris-Telehouse, Lyon-IX, and Marseille-CRS).
  • CGNAT Characteristics: Orange implements large-scale NAT444 architecture across its mobile operations. Public egress ranges often present as /18 or /17 blocks. The IP churn rate on Orange 5G is tied to radio bearer reconfiguration: maintaining a data session without an airplane-mode reset can preserve an IP for hours, but triggering an industrial AT command or network disconnect forces a transition to a new public address within the AS3215 allocation.
  • Reputation Metrics: Orange holds the highest trust rating among global fraud engines. Traffic originating from Orange mobile subnets rarely triggers deep browser integrity checks unless client-side JavaScript execution actively reports automated behaviors.

SFR / Altice France (AS15557)

SFR manages one of France's densest LTE-Advanced and 5G networks, operating under ASN 15557 (Société Française du Radiotéléphone).

  • Routing Architecture: SFR routes significant portions of its mobile traffic through regional concentrators located in the Paris Île-de-France and Lyon metropolitan areas.
  • CGNAT Characteristics: SFR assigns dynamic IPv4 addresses from dynamic enterprise and consumer pools. Port allocation per subscriber is strictly managed via dynamic Port Control Protocol (PCP) or deterministic CGNAT, resulting in high subscriber density per public IP.
  • Operational Utility: SFR's mobile IPs demonstrate exceptional longevity for long-polling tasks, making them optimal for tracking volatile inventory shifts, live stock drops, and high-frequency SERP adjustments.

Free Mobile (AS12322)

Owned by Iliad, Free Mobile revolutionized the French market with aggressive infrastructure expansion.

  • IPv6 and 464XLAT Transition: Free Mobile is an industry leader in IPv6 deployment. In their 5G and 4G networks, Free heavily utilizes 464XLAT (RFC 6877) architecture. The client device receives an IPv6-only bearer interface, and an on-device Customer-Side Translator (CLAT) translates IPv4 packets into IPv6, which traverse the core network to a Provider-Side Translator (PLAT / NAT64) before exiting to the IPv4 internet.
  • Subscriber Multiplexing Ratio: Free Mobile's CGNAT sharing ratio is exceptionally high. Tens of thousands of real mobile users may egress through a narrow /20 block of AS12322 addresses. Consequently, anti-bot engines treat Free Mobile's IPv4 ranges with wide tolerances to avoid triggering catastrophic false-positive lockouts on consumer users.

Bouygues Telecom (AS5410)

Bouygues Telecom operates AS5410, providing widespread geographic coverage and enterprise-grade cellular transit.

  • Characteristics: Bouygues maintains distinct separation between its fixed broadband lines (Bbox) and its mobile cellular gateways. The mobile IP pools on AS5410 are characterized by clean historical routing reputations, with minimal abuse records on public blocklists (Spamhaus, Barracuda, Project Honey Pot).
  • Target Suitability: Bouygues mobile lines are particularly effective against Vinted and Fnac, where secondary account generation and marketplace API extraction require clean connection handshakes.

4. Hardware-Level Proxy Architecture: Industrial Teltonika RUTX50 / TRB500

Many retail proxy services assemble "mobile proxy" farms using compromised consumer smartphones (Android devices connected over Wi-Fi, sharing data via USB tethering) or cheap consumer LTE dongles (Huawei E3372). In mission-critical enterprise data operations, these consumer setups fail due to thermal throttling, battery bloat, unstable USB serial buses, and uncontrolled OS-level connection interference.

Industrial proxy deployments rely on dedicated field-grade cellular hardware: the Teltonika RUTX50 (dual-SIM 5G router) or the Teltonika TRB500 (compact industrial 5G gateway).

INDUSTRIAL HARDFORMAT CELLULAR ARCHITECTURE (PROXYM):

  ┌─────────────────────────────────────────────────────────────┐
  │                 Proxym Industrial Rack System               │
  │                                                             │
  │   [ Dedicated 5G Enterprise SIM: Orange / SFR / Free ]      │
  │                               │                             │
  │                               ▼                             │
  │        [ Teltonika RUTX50 / TRB500 Industrial Gateway ]     │
  │        ├─ Quectel 5G Sub-6GHz Module (Up to 3.3 Gbps)       │
  │        ├─ Quad-Core ARM Cortex-A7 CPU                       │
  │        ├─ RutOS (Hardened OpenWrt Kernel 5.4+)              │
  │        ├─ Physical Ethernet Egress (Gigabit RJ-45)          │
  │        └─ Microcontroller-driven AT Command Control         │
  └──────────────────────────────┬──────────────────────────────┘
                                 │ Direct Cat6 Shielded LAN
                                 ▼
         [ Private Out-of-Band Management & Proxy Core ]
         ├─ SOCKS5 Authentication (RFC 1928)
         ├─ Encrypted HTTP/HTTPS Connect Tunneling
         └─ RESTful API Endpoint (Instant IP Rotation via AT+CFUN)

The Engineering Advantage of Industrial Hardware

  • Thermal Stability and Continual MTBF: Consumer phones experience thermal throttling under heavy cryptographic proxy loads (TLS termination, SOCKS5 multiplexing), causing connection drops and radio resets. The Teltonika RUTX50 features a ruggedized aluminum housing designed for passive heat dissipation, capable of sustained multi-gigabit throughput in operating temperatures from -40°C to 75°C.
  • Physical 5G Sub-6GHz Performance: Utilizing multi-band Quectel 5G modems, these industrial devices support 4x4 MIMO configurations, aggregation across multiple 5G bands (n1, n3, n7, n28, n78), and maximum theoretical downlinks of 3.3 Gbps. This eliminates proxy-induced latency bottlenecks during high-concurrency scraping operations.
  • Deterministic Programmatic IP Rotation via AT Commands: Consumer Android proxies rotate IP addresses by toggling the Android UI "Airplane Mode" using ADB (Android Debug Bridge), a brittle mechanism prone to OS crashes. The Teltonika platform executes low-level Hayes AT commands directed to the modem firmware over an internal serial bus (/dev/ttyUSB2 or via proprietary gsmctl binaries).
# Executing instantaneous IP rotation via Teltonika CLI / API
# Disables cellular RF circuit (Airplane Mode Equivalent)
gsmctl -A 'AT+CFUN=4'

# Wait 2000ms for carrier detachment
sleep 2

# Enables cellular RF circuit (Re-attaches to 5G-NR Base Station)
# Forces GTP-U tunnel tear-down and initiates new CGNAT IP allocation
gsmctl -A 'AT+CFUN=1'

This cycle detaches the modem from the eNodeB/gNodeB and re-initiates the 3GPP attach procedure, forcing the French carrier's PGW/UPF to provision a fresh, unrelated public CGNAT IPv4 address in 4 to 8 seconds.

5. How Anti-Fraud Engines Score Mobile Carrier Ranges vs. Foreign Proxies

Modern anti-fraud systems (Sift, Riskified, ThreatMetrix, DataDome) quantify trust using continuous Bayesian scoring models.

Every incoming HTTP/HTTPS connection is evaluated through a composite probability function:

$$P(\text{Automated} \mid \vec{x}) = \frac{P(\vec{x} \mid \text{Automated}) P(\text{Automated})}{P(\vec{x})}$$

Where $\vec{x}$ is a vector of observed network, transport, and application-layer metrics.

+───────────────────────────+───────────────────────────+───────────────────────────+
| Evaluated Vector          | Datacenter / Foreign VPN  | French Industrial 5G      |
|                           | (OVH / Hetzner / AWS)     | (Orange / Free / SFR)     |
+───────────────────────────+───────────────────────────+───────────────────────────+
| BGP ASN Routing Type      | Hosting / Data Center     | Mobile / Cellular (MNO)   |
| IP Reputation Score       | Low (0.05 - 0.20)         | High (0.92 - 0.99)        |
| TCP SYN Packet MSS        | 1460 (Standard Ethernet)  | 1360 - 1400 (3GPP GTP)    |
| p0f OS vs UA Fingerprint  | Linux (Host) != Mac (UA)  | True Cellular Match       |
| Round-Trip Time (RTT)     | Sub-10ms or >150ms cross  | 25ms - 55ms (Local 5G)    |
| Port Scan Fingerprint     | Open 22, 80, 443, 8080    | Fully Closed / Filtered   |
+───────────────────────────+───────────────────────────+───────────────────────────+

TCP/IP Stack Fingerprinting (p0f and SYN Analysis)

Anti-bot edge nodes do not trust the HTTP User-Agent header; they cross-examine it against Layer 4 transport metrics.

  1. Maximum Segment Size (MSS): Standard Ethernet networks operate with an MTU of 1500 bytes, yielding an MSS of 1460 bytes ($1500 - 20 \text{ (IP)} - 20 \text{ (TCP)}$). Cellular interfaces, however, introduce encapsulation overhead (GTP-U, PDCP). MNOs clamp the TCP MSS to values between 1360 and 1400 bytes. If a request advertises an iPhone Safari User-Agent but presents an MSS of 1460 on an OVH IP, the system flags the connection as an emulated client routing through a datacenter.
  2. TCP Window Size and Flags: Linux kernels (used on 99% of datacenter proxy servers) construct SYN packets with specific initial window sizes, scale factors, and option orders (e.g., MSS,SackOK,TS,NOP,WScale). Windows and macOS/iOS devices exhibit entirely different ordering and window size values.
  3. Passive OS Fingerprinting: When an automated script claims to be an Apple iPad running iOS over Safari, but the p0f Layer 4 signature detects an underlying Linux 5.15 kernel (the host server running the VPN exit node), the request's anomaly score jumps to maximum, triggering an automated challenge.

Round-Trip Time (RTT) Geometry

Advanced anti-fraud engines measure the TCP handshake delta:

$$\text{RTT}_{\text{Handshake}} = T_{\text{ACK}} - T_{\text{SYN-ACK}}$$

If an automated user targets leboncoin.fr (servers hosted primarily around Paris) via an exit node claiming to be in France, but the physical TCP handshake completes in under 1.2 milliseconds, the connection is instantly flagged: only a server located in the exact same colocation facility or campus could achieve this latency.

Real French 5G connections over Orange, SFR, or Free display natural radio interface propagation latencies between 22ms and 65ms, directly matching real human device profiles.

6. Strategic Cost Analysis: Dedicated Hardware vs. "Rotating" Pools

Data operations teams regularly balance two architectures: cheap, metered "residential" proxy pools and physical, dedicated industrial 4G/5G mobile ports.

METRIC BREAKDOWN: DEDICATED 5G PORT VS. RESIDENTIAL POOLS

   DEDICATED 5G PORT (Proxym Model)
   [ Dedicated Physical SIM ] ──> Flat Fee (€70 - €80/month)
                                 ├── 200 GB Fair Use included
                                 ├── Cost per GB: ~€0.40/GB
                                 ├── Dedicated bandwidth (no sharing)
                                 └── Real-time hardware control (AT commands)

   ROTATING RESIDENTIAL POOLS (Standard SaaS Model)
   [ Shared P2P Network ]    ──> Pay-Per-Gigabyte (€8 - €15/GB)
                                 ├── Cost for 200 GB: €1,600 - €3,000/month
                                 ├── High connection drops (peer disconnects)
                                 ├── Dynamic IP leakage / Dirty peer IPs
                                 └── Zero control over physical hardware

The Financial Multiplier of Data-Heavy Automation

Modern web properties deliver rich web assets: dynamic React bundles, tracking scripts, web fonts, and high-resolution webp images.

  • Average page weight on French retail platforms (Fnac, Darty, Leboncoin) ranges from 2.5 MB to 6.8 MB per full render.
  • Scraping 500,000 product pages per month via a metered residential proxy service at €10.00 per GB leads to massive operational expenses:

$$\text{Data Volume} = 500,000 \times 4.0\text{ MB} = 2,000,000\text{ MB} = 2,000\text{ GB}$$

$$\text{Metered Residential Cost} = 2,000\text{ GB} \times €10.00/\text{GB} = \mathbf{€20,000/\text{month}}$$

In contrast, deploying dedicated industrial cellular hardware from Proxym operates under an enterprise flat-rate model:

  • Single Port: €80 / month.
  • Multi-Port (3+ Ports): €70 / port / month.
  • Fair-Use Ceiling: 200 GB high-speed enterprise data included per port (yielding an effective base cost of €0.40 per GB).
  • Trial Evaluation: Proxym offers a full test setup at €5 for the first month utilizing promo code DECOUVERTE5.

For enterprise operations scraping terabytes of localized competitive intelligence, transitioning from metered residential networks to dedicated industrial 5G modems cuts infrastructure costs by 85% to 95% while dramatically improving connection stability.

7. Technical Implementation: Localized SERP & E-Commerce Playbook

The following engineering blueprints demonstrate how to execute high-throughput extraction pipelines utilizing localized French mobile proxies.

Playbook 1: Scraping Localized Google SERPs (Paris vs. Lyon)

Google serves distinct search rankings based on geolocation. In the absence of native browser geolocation permissions, Google relies on:

  1. The incoming IPv4/IPv6 BGP routing origin and local carrier edge point.
  2. The uule parameter (canonical base64-encoded geographic location parameter).

When you pair a French carrier proxy (Orange or Free) with matching French UULE strings, Google processes the connection through domestic consumer nodes, preventing localized blocking.

GOOGLE SERP EXTRACTION PIPELINE:

[ Python Playwright Client ]
             │
             ├── Set Viewport, User-Agent, Mobile Emulation
             ├── Construct UULE for Target French City (e.g., Paris)
             │
             ▼
[ Proxym SOCKS5/HTTP Gateway ] ──> [ Dedicated Teltonika RUTX50 5G Modem ]
                                               │
                                               ▼ (French MNO Carrier Core: AS3215)
                                     [ Google Search Node (France) ]
                                               │
                                               ▼
                              [ Localized French SERP: 200 OK ]

Production Python (Playwright) Localized Search Extraction

#!/usr/bin/env python3
"""
Production Google SERP Scraper configured for French Mobile Proxies (Proxym).
Extracts hyper-localized search engine results without datacenter blocks.
"""

import asyncio
import base64
import urllib.parse
from playwright.async_api import async_playwright

# Proxym Endpoint Configuration
PROXY_HOST = "eu.proxym.io"
PROXY_PORT = "1080"  # Dedicated SOCKS5/HTTP Port
PROXY_USER = "proxym_client_1042"
PROXY_PASS = "secure_carrier_auth_token"

def generate_uule(city_name: str) -> str:
    """
    Generates a Google canonical UULE parameter for geographic targeting.
    Example: 'Paris,Ile-de-France,France' -> '&uule=a+...'
    """
    secret_key = len(city_name)
    secret_char = chr(secret_key)
    base64_encoded = base64.b64encode(city_name.encode('utf-8')).decode('utf-8')
    # Google uses a specific base64 encoding scheme where spaces become underscores/hyphens
    clean_base64 = base64_encoded.replace('+', '-').replace('/', '_').rstrip('=')
    return f"a+{secret_char}{clean_base64}"

async def scrape_french_serp(query: str, location_canonical: str):
    uule_code = generate_uule(location_canonical)
    target_url = (
        f"https://www.google.fr/search?q={urllib.parse.quote(query)}"
        f"&hl=fr&gl=fr&uule={uule_code}"
    )

    async with async_playwright() as p:
        # Launch Chromium with anti-detection flags matching real mobile behavior
        browser = await p.chromium.launch(
            headless=True,
            proxy={
                "server": f"http://{PROXY