Skip to content

antibot-benchmark.md

Latest commit

 

History

History
322 lines (254 loc) · 10 KB

antibot-benchmark.md

File metadata and controls

322 lines (254 loc) · 10 KB
Raw

Antibot Systems Benchmark

Gakido includes a comprehensive benchmark tool to test browser impersonation against various antibot detection systems.

Quick Start

# Run with default profiles (13 profiles)
uv run python examples/antibot_benchmark.py

# Test specific profile
uv run python examples/antibot_benchmark.py --profile chrome_144

# Test all 24 base profiles
uv run python examples/antibot_benchmark.py --all-profiles

Test Categories

The benchmark tests 9 endpoints across 4 categories:

1. TLS Fingerprint Tests

Endpoint URL Description
tls_browserleaks tls.browserleaks.com JA3/JA4 fingerprint analysis
scrapfly_ja3 tools.scrapfly.io Scrapfly JA3 fingerprint

2. HTTP Header Analysis

Endpoint URL Description
httpbin_headers httpbin.org/headers HTTP header echo service
httpbin_user_agent httpbin.org/user-agent User-Agent verification

3. Bot Detection Services

Endpoint URL Description
nowsecure nowsecure.nl NowSecure bot detection test
sannysoft bot.sannysoft.com Sannysoft bot detection
incolumitas bot.incolumitas.com Incolumitas bot detection

4. Cloudflare Endpoints

Endpoint URL Description
cloudflare_trace cloudflare.com/cdn-cgi/trace Cloudflare trace endpoint
cloudflare_1111 1.1.1.1/cdn-cgi/trace Cloudflare 1.1.1.1 trace

Usage

Basic Usage

# Default profiles
uv run python examples/antibot_benchmark.py

# Single profile
uv run python examples/antibot_benchmark.py --profile chrome_144

# Multiple profiles
uv run python examples/antibot_benchmark.py -p chrome_144 -p firefox_147 -p brave_131

# All base profiles (24 profiles)
uv run python examples/antibot_benchmark.py --all-profiles

Options

Option Description
--profile, -p Profile to test (can be specified multiple times)
--all-profiles Test all 24 base profiles
--async Use async client for testing
--timeout, -t Request timeout in seconds (default: 15)
--output, -o Save results to JSON file
--quiet, -q Suppress verbose output

Examples

# Async mode (faster for multiple profiles)
uv run python examples/antibot_benchmark.py --async

# Custom timeout
uv run python examples/antibot_benchmark.py --timeout 30

# Save results to JSON
uv run python examples/antibot_benchmark.py --output results.json

# Quiet mode with JSON output
uv run python examples/antibot_benchmark.py -q -o results.json

Default Profiles Tested

The benchmark tests these 13 profiles by default:

Profile Platform Client Hints
chrome_120 macOS ✅
chrome_144 macOS ✅
chrome_131_windows Windows ✅
chrome_136_linux Linux ✅
firefox_133 macOS ❌
firefox_147 Linux ❌
safari_184_macos macOS ❌
safari_187_ios iOS ❌
edge_131 Windows ✅
edge_144 Windows ✅
brave_131 Windows ✅
opera_115 Windows ✅
tor_145 Windows ❌

Sample Output

======================================================================
  GAKIDO ANTIBOT SYSTEMS BENCHMARK
======================================================================
Profiles to test: 4
Timeout: 10.0s

============================================================
Testing profile: chrome_144
============================================================
  ✅ tls_browserleaks          HTTP 200   711ms
  ✅ scrapfly_ja3              HTTP 200   950ms
  ✅ httpbin_headers           HTTP 200   873ms
  ✅ httpbin_user_agent        HTTP 200   260ms
  ✅ nowsecure                 HTTP 200   2389ms
  ✅ sannysoft                 HTTP 200   608ms
  ✅ incolumitas               HTTP 200   1087ms
  ✅ cloudflare_trace          HTTP 200   469ms
  ✅ cloudflare_1111           HTTP 200   419ms

  Summary: 9/9 passed (100.0%) | Avg: 895ms

======================================================================
  BENCHMARK SUMMARY
======================================================================

Profiles tested: 4
Total tests run: 36
Overall success rate: 100.0%

--- Category Breakdown ---
  TLS Fingerprint     : 8/8 (100.0%)
  HTTP Headers        : 8/8 (100.0%)
  Bot Detection       : 12/12 (100.0%)
  Cloudflare          : 8/8 (100.0%)

--- Profile Rankings ---
   1. chrome_144                100.0% | hints:✓ | macOS
   2. firefox_147               100.0% | hints:✗ | Linux
   3. safari_184_macos          100.0% | hints:✗ | macOS
   4. brave_131                 100.0% | hints:✓ | Windows

🏆 Perfect scores: chrome_144, firefox_147, safari_184_macos, brave_131

JSON Output Format

When using --output, results are saved in this format:

{
  "timestamp": "2025-01-25T16:15:00.000000",
  "gakido_version": "0.1.0",
  "total_profiles_tested": 4,
  "total_tests_run": 36,
  "overall_success_rate": 100.0,
  "profiles": [
    {
      "profile_name": "chrome_144",
      "platform": "macOS",
      "has_client_hints": true,
      "has_canvas_webgl": true,
      "user_agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7)...",
      "tests": [
        {
          "endpoint_name": "tls_browserleaks",
          "url": "https://tls.browserleaks.com/json",
          "description": "TLS/JA3 fingerprint analysis",
          "success": true,
          "status_code": 200,
          "response_time_ms": 711.5,
          "error": null,
          "details": {"response": {"ja3_hash": "..."}}
        }
      ],
      "total_tests": 9,
      "passed_tests": 9,
      "failed_tests": 0,
      "success_rate": 100.0,
      "avg_response_time_ms": 895.2
    }
  ],
  "summary": {
    "category_stats": {
      "TLS Fingerprint": {"total": 8, "passed": 8},
      "HTTP Headers": {"total": 8, "passed": 8},
      "Bot Detection": {"total": 12, "passed": 12},
      "Cloudflare": {"total": 8, "passed": 8}
    },
    "best_profile": "chrome_144",
    "worst_profile": "tor_145",
    "profiles_with_100_percent": ["chrome_144", "firefox_147"]
  }
}

Programmatic Usage

from examples.antibot_benchmark import AntibotBenchmark, AsyncAntibotBenchmark
import asyncio

# Sync benchmark
benchmark = AntibotBenchmark(
    profiles=["chrome_144", "firefox_147"],
    timeout=15.0,
    verbose=True,
)
report = benchmark.run()

# Access results
print(f"Success rate: {report.overall_success_rate}%")
for profile in report.profiles:
    print(f"{profile.profile_name}: {profile.success_rate}%")

# Save to JSON
benchmark.save_report(report, "results.json")

# Async benchmark
async def run_async():
    benchmark = AsyncAntibotBenchmark(
        profiles=["chrome_144", "firefox_147"],
        timeout=15.0,
    )
    return await benchmark.run_async()

report = asyncio.run(run_async())

Adding Custom Endpoints

You can extend the benchmark by adding custom endpoints:

from examples.antibot_benchmark import AntibotBenchmark, BOT_DETECTION_ENDPOINTS

# Add custom endpoint
BOT_DETECTION_ENDPOINTS["my_endpoint"] = {
    "url": "https://example.com/bot-check",
    "description": "My custom bot check",
    "check": lambda r: r.get("is_bot") == False,  # Custom validation
    "check_status": lambda s: s == 200,
}

# Run benchmark with custom endpoint included
benchmark = AntibotBenchmark(profiles=["chrome_144"])
report = benchmark.run()

Interpreting Results

Success Criteria

  • TLS Fingerprint: Returns valid JA3/JA4 hash
  • HTTP Headers: Returns expected header structure
  • Bot Detection: HTTP 200 response (page loads without block)
  • Cloudflare: Contains fl= in trace response

What Affects Results

Factor Impact
TLS Configuration JA3/JA4 fingerprint matching
HTTP/2 Settings SETTINGS frame fingerprint
Header Order Header ordering detection
Client Hints Sec-CH-UA consistency
User-Agent Browser identification

Recommendations

  • Use Chrome profiles for best compatibility
  • Enable client hints for Chromium-based detection
  • Use platform-specific profiles (e.g., chrome_131_windows for Windows targets)
  • Rotate profiles to avoid fingerprint-based blocking

Results: fingerprint comparison

A direct comparison against tls.peet.ws with the chrome_120 profile. Every client sends the same Chrome User-Agent, so the only variable is the TLS / HTTP/2 fingerprint.

Client JA4 (TLS) HTTP/2 (Akamai)
requests t13d1812h1_85036bcba153_… — (HTTP/1.1)
httpx t13d1812h1_85036bcba153_… — (HTTP/1.1)
gakido (tls_backend="stdlib") t13d3112h2_e8f1e7e78f70_… 1:65536;2:0;4:6291456;5:16384;8:0;6:262144|15663105|0|m,a,s,p
gakido (native) t13d1516h2_8daaf6152771_02713d6af862 1:65536;2:0;4:6291456;6:262144|15663105|0|m,a,s,p
real Chrome 120 t13d1516h2_8daaf6152771_02713d6af862 1:65536;2:0;4:6291456;6:262144|15663105|0|m,a,s,p
  • requests / httpx present an OpenSSL fingerprint over HTTP/1.1 — nothing like a browser.
  • gakido's pure-Python path gets the HTTP/2 fingerprint almost right (the extra 5:16384;8:0 is forced by the h2 library) but TLS is still OpenSSL.
  • gakido's native backend matches real Chrome on both JA4 and the Akamai HTTP/2 fingerprint, byte for byte. For Chromium profiles the extension order is randomized per connection, so the JA3 hash also varies the way Chrome's does.

What this does and does not mean

Matching the fingerprint defeats anti-bot systems that block on the TLS / HTTP fingerprint itself — a large and common class. It does not solve JavaScript challenges such as Cloudflare's managed challenge or Turnstile: those require executing JavaScript and cannot be passed by any raw HTTP client (gakido, curl_cffi, or otherwise) without a real/headless browser. Treat gakido as the transport-fingerprint layer, not a challenge solver.