Skip to content
Bekbolot Bekbolot
December 15, 2025 ยท 5 min read

Release of ADA Detection v1

ADA Detection v1

We are announcing the release of ada_detection_v1, a specialized challenge designed to evaluate detection capabilities within anti-detect browser environments. Unlike previous challenges that focused on standard driver fingerprints, ADA operates entirely inside NST-Browser, where traditional static signals are masked.

The goal of this challenge is to push the boundaries of behavioral analysis, requiring solutions that detect how automation moves and acts, rather than what it claims to be.

Release Date

ADA Detection v1 is live starting December 15, 2025 at 10:00 UTC. All new miner submissions are evaluated using this version.

What is New in ADA Detection v1

This release shifts the focus from static property inspection to dynamic behavioral analysis in a hardened environment.

The Anti-Detect Environment

ADA Detection v1 runs all evaluations inside NST-Browser. In this environment:

  • Fingerprints are masked: Standard "bot" signals are actively spoofed.
  • Profiles are fresh: Every run starts with a clean profile.
  • Isolation: There is zero shared state between runs.

This simulates sophisticated, real-world scraping operations where "navigator.webdriver" checks are obsolete. Participants must rely on orchestration timing, WebSocket control patterns, and behavioral anomalies.

Strict Human Safety

This challenge introduces a zero-tolerance policy for false positives. Human Safety is the kill switch.

  • Real human interactions are randomly injected into the evaluation flow.
  • If your solution flags more than 2 humans as bots, your entire score for the submission is 0.0.

This ensures that high-performing detection scripts are safe to deploy in production environments without disrupting real user traffic.

Targeted Framework Coverage

Miners must submit distinct detection logic for four specific automation frameworks commonly used with anti-detect browsers:

  • automation
  • nodriver
  • playwright
  • patchright
  • puppeteer

Missing detection logic for any of these frameworks invalidates the submission.

"Payload-Based" Detection Flow

Unlike previous iterations that might scan for file artifacts, ADA v1 uses a runtime payload system:

  1. Scripts execute in the browser page.
  2. When automation is confirmed, scripts send a payload to a local /_payload endpoint.
  3. Silence is required during human sessions. Any payload sent during a human interaction counts as a critical mistake.

Submission Structure

We have streamlined the submission format to be strictly modular. Submissions must be a JSON object containing exactly four files, named precisely:

  {
    "detection_files": [
      { "file_name": "automation.js", "content": "..." },
      { "file_name": "nodriver.js", "content": "..." },
      { "file_name": "playwright.js", "content": "..." },
      { "file_name": "patchright.js", "content": "..." },
      { "file_name": "puppeteer.js", "content": "..." }
    ]
  }
  • File names must match exactly.
  • Each file is responsible only for detecting its assigned framework.
  • Logic must be self-contained JavaScript (ES6+).

Updated Scoring Logic

Scoring in ADA v1 is a composite metric normalized between 0.0 and 1.0.

1. The Human Safety Gate

Before any points are awarded, human accuracy is checked.

  • Start: 1.0 points.
  • Penalty: -0.1 per mistake.
  • Limit: > 2 mistakes = 0.0 Final Score (Immediate Failure).

2. Framework Detection (All-or-Nothing)

For each framework (e.g., nodriver), your script is tested against multiple runs.

  • 1 Point: Perfect detection across all runs for that framework.
  • 0 Points: Any missed detection or collision (reporting the wrong framework).

3. Normalization

The final score is calculated by summing your Human Accuracy, Automation Accuracy, and Framework Points, then normalizing against the total number of test cases.

\[ \text{Final Score} = \frac{\text{Human Score} + \text{Automation Score} + \text{Framework Points}}{\text{Total Frameworks} + 1 \text{ Human} + 1 \text{ Auto}} \]

Similarity & Incentives

To encourage continuous innovation:

  • Similarity Check: Submissions are compared against historical data. High similarity to previous solutions results in penalties or rejection.
  • Time Decay: Scores naturally decay over a 15-day period, incentivizing miners to regularly update and refine their detection heuristics.

Summary

ADA Detection v1 represents a paradigm shift:

  • Environment: NST-Browser (Anti-Detect).
  • Philosophy: Behavioral over Static.
  • Safety: Strict penalties for false positives.
  • Format: Modular, payload-based detection.

This challenge rewards miners who can solve the hardest problem in bot detection: distinguishing a sophisticated script from a real human without relying on easily spoofed signals.

Testing Dependency

Miner need to buy profession plan of nstbrowser for testing challenge locally. Miners can buy it from NstBrowser's official page