New challenge & improve stability

.github/workflows/1.bump-version.yml

@@ -17,7 +17,7 @@ jobs:
       contents: write
     steps:
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
       - name: Bump version
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/2.build-publish.yml

@@ -17,9 +17,9 @@ jobs:
   #     contents: read
   #   steps:
   #     - name: Checkout
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v6
   #     - name: Set up Python
-  #       uses: actions/setup-python@v5
+  #       uses: actions/setup-python@v6
   #       with:
   #         python-version: "3.10"
   #     - name: Install dependencies
@@ -37,13 +37,13 @@ jobs:
       contents: write
     steps:
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
       - name: Get latest version
         run: |
           git pull origin main
           echo "Current version: v$(./scripts/get-version.sh)"
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v6
         with:
           python-version: "3.10"
       - name: Install dependencies

.github/workflows/3.update-changelog.yml

@@ -14,7 +14,7 @@ jobs:
       contents: write
     steps:
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
       - name: Update changelog
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/publish-docs.yml

@@ -17,11 +17,11 @@ jobs:
       contents: write
     steps:
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v6
         with:
           python-version: "3.10"
       - name: Install dependencies

docs/challenges/ada_detection/.nav.yml

@@ -1,4 +1,4 @@
 nav:
   - README.md
-  - Anti-Detect Automation Detection v1: v1.md
+  - Anti-Detect Automation Detection v2: v2.md
   - "*"

docs/challenges/ada_detection/README.md

@@ -2,7 +2,7 @@
 
 ## Overview
 
-The **Anti-Detect Automation Detection (AAD)** challenge evaluates a participant`s ability to reliably detect browser automation frameworks operating inside anti-detect browsers, while preserving human safety.
+The **Anti-Detect Automation Detection (AAD)** challenge evaluates a participant's ability to reliably detect browser automation frameworks operating inside anti-detect browsers, while preserving human safety.
 
 Evaluation runs simulate real-world anti-detect usage where static signals are masked and fingerprints are fresh. Detection must rely on runtime behavior and orchestration patterns.
 
@@ -17,23 +17,23 @@ Each evaluation run involves:
 !!! Info "NST-Browser Dependency"
     Participants need an API key from the [NSTBrowser](https://www.nstbrowser.io/en/pricing) dashboard (Professional plan recommended) for local testing.
 
-## Features of ADA Detection v1
+## Features of ADA Detection v2
 
-- **Behavior-first detection**: Focus on runtime behavioral analysis inside NST-Browser; static driver fingerprints are spoofed.
-- **Payload-based flow**: Detection scripts run in-page, send payloads to local `/_payload` when automation is confirmed, and must stay silent during human sessions.
-- **Human safety gate**: Random human interactions are injected; flagging more than 2 humans as bots zeros the submission.
-- **Target coverage update**: Required detectors include `automation`, `nodriver`, `playwright`, `patchright`, and `puppeteer`.
-- **Fresh profiles, no state**: Each run uses a fresh profile with zero shared state between runs.
+- **Hardened NSTBrowser Environment**: Advanced obfuscation including dynamic fingerprinting and hardware simulation (8GB RAM, 16 Cores).
+- **Engine-Level Stealth**: `AutomationControlled` flags are disabled at the browser engine level to eliminate static detection signals.
+- **Fail-Fast Scoring**: High-stakes evaluation where missing critical targets (Selenium) or exceeding miss limits results in an immediate 0.0 score.
+- **Protocol-Level Verification**: Mandatory detection of low-level communication patterns (Webdriver and WebSocket).
+- **Selenium Safety Gate**: Mandatory detection of `seleniumbase` and `selenium_driverless` is required for incentive eligibility.
 - **Similarity/time decay**: Similarity penalties apply to lookalike submissions; scores decay over 15 days to encourage refreshed heuristics.
 
 ## Evaluation Flow
 
 1. **Submission Received**: Detection scripts are submitted via the `/score` endpoint.
 2. **Task Generation**: A randomized sequence of multiple automation framework runs and human interactions is generated.
 3. **NST-Browser Launch**: A clean instance is started for each task.
-4. **Execution**: Automation frameworks connect via WebSocket, while humans interact manually.
+4. **Execution**: Automation frameworks connect via WebSocket or WebDriver, while humans interact manually.
 5. **Detection Phase**: Scripts may emit detection payloads to `/_payload`.
-6. **Scoring**: Results are aggregated and normalized.
+6. **Scoring**: Results are aggregated and normalized using the Fail-Fast pillars.
 
 ## Technical Constraints
 
@@ -59,11 +59,16 @@ Submit your detection scripts as a JSON payload with the following structure:
 ```json
 {
   "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_name": "puppeteer.js", "content": "..." },
+    { "file_name": "puppeteer_extra.js", "content": "..." },
+    { "file_name": "zendriver.js", "content": "..." },
+    { "file_name": "selenium_driverless.js", "content": "..." },
+    { "file_name": "seleniumbase.js", "content": "..." },
+    { "file_name": "webdriver.js", "content": "..." },
+    { "file_name": "websocket.js", "content": "..." }
   ]
 }
 ```
@@ -72,7 +77,8 @@ Each file must be named exactly as shown and contain self-contained JavaScript (
 
 ## Challenge Versions
 
-- [**v1** (Active after Dec 15, 2025 10:00 UTC)](./v1.md) - Payload-based detection with human safety gate
+- [**v2** (Active after Feb 14, 2026 14:00 UTC)](./v2.md) - Hardened detection with Fail-Fast scoring
+- [**v1** (Deprecated)](./depricated/v1.md)
 
 ## Resources & Guides
 

docs/challenges/ada_detection/v2.md

@@ -0,0 +1,145 @@
+---
+title: Anti-Detect Automation Detection v2
+---
+
+# Anti-Detect Automation Detection v2 (Active after 2026 February 14 14:00 UTC)
+
+## Overview
+
+**ADA Detection v2** focuses on advanced behavioral detection inside **NSTBrowser** environments. This version introduces hardened detection requirements where static signals are masked and engine-level obfuscation is active.
+
+### Browser Hardening Features
+
+- **Dynamic Fingerprinting**: Every session generates unique, randomized browser identifiers to prevent static pattern matching.
+- **Hardware Simulation**: Realistic signals are injected for `deviceMemory` (8GB), `hardwareConcurrency` (16), and screen resolution (1280x1024).
+- **Engine-Level Stealth**: Disables `AutomationControlled` flags at the browser engine level, eliminating traditional automation leaks.
+
+Scoring has transitioned to a **Fail-Fast** model, emphasizing accuracy across three critical pillars: Human Detection, Framework Detection (with a Selenium Safety Gate), and Protocol Accuracy.
+
+For general challenge information, environment details, and plagiarism policies, please refer to the [AAD README](./README.md).
+
+---
+
+## Target Frameworks & Protocols
+
+Participants must submit detection scripts for the following frameworks and protocols:
+
+- **Frameworks**: `nodriver`, `playwright`, `patchright`, `puppeteer`, `puppeteer_extra`, `zendriver`, `selenium_driverless`, `seleniumbase`
+- **Protocols**: `webdriver`, `websocket`
+
+Missing any of these scripts results in an invalid submission.
+
+### Selenium Safety Gate
+
+Missing detection for either `seleniumbase` or `selenium_driverless` results in an immediate **final score of 0.0**. These frameworks are the primary automation vectors in v2; failure to detect them indicates a fundamental gap in the detection logic.
+
+---
+
+## Submission Format
+
+Submissions must follow this structure:
+
+```json
+{
+  "detection_files": [
+    { "file_name": "nodriver.js", "content": "/* logic */" },
+    { "file_name": "playwright.js", "content": "/* logic */" },
+    { "file_name": "patchright.js", "content": "/* logic */" },
+    { "file_name": "puppeteer.js", "content": "/* logic */" },
+    { "file_name": "puppeteer_extra.js", "content": "/* logic */" },
+    { "file_name": "zendriver.js", "content": "/* logic */" },
+    { "file_name": "selenium_driverless.js", "content": "/* logic */" },
+    { "file_name": "seleniumbase.js", "content": "/* logic */" },
+    { "file_name": "webdriver.js", "content": "/* logic */" },
+    { "file_name": "websocket.js", "content": "/* logic */" }
+  ]
+}
+```
+
+### Rules
+
+- File names must match the framework/protocol names exactly.
+- Each file detects **only its own framework or protocol**.
+- No extra files or outputs are allowed.
+
+---
+
+## Scoring System: The Three Pillars
+
+ADA v2 uses a **Fail-Fast** scoring model. If a submission fails any of the three critical pillars, the final score is immediately set to **0.0**.
+
+### 1. Human Detection
+
+Miners must distinguish between automated tasks and human-injected sessions.
+
+- **Limit**: You are allowed a maximum of **1 mistake**. Exceeding this limit results in an immediate **score of 0.0**.
+- **Weight**: Perfect detection grants 1.0 point for this pillar. Partial penalties apply for a single miss based on the ratio of human injections.
+
+### 2. Framework Detection
+
+Points are earned for correctly identifying the specific automation framework.
+
+- **Selenium Gate**: Missing `seleniumbase` or `selenium_driverless` zeros the entire score.
+- **Density**: You earn **1.0 point** for a framework only if you detect it perfectly in **all 3 of its runs**.
+- **Collision**: Reporting more than one framework or an incorrect protocol for a given session results in a collision penalty (earning only 0.1 points instead of 1.0).
+
+### 3. Protocol Accuracy
+
+Validates the low-level communication patterns of the browser.
+
+- **Webdriver Protocol**: Expected to be `true` for Selenium-based frameworks and `false` for others (human, playwright, etc.).
+- **Websocket Protocol**: Expected to be `true` for non-Selenium frameworks (playwright, puppeteer, etc.) and `false` for human sessions and Selenium.
+- **Threshold**: You are allowed a maximum of **1 miss** per protocol type. Exceeding this limit results in an immediate **score of 0.0**.
+
+### Final Formula
+
+The final score is normalized between 0.0 and 1.0 using the formula:
+
+```math
+Final Score = (Human Score + Framework Score + Protocol Score) / 10
+```
+
+__(Where 10 = 8 frameworks + 1 human pillar + 1 protocol pillar)__
+
+---
+
+## Incentive Eligibility
+
+To maintain high detection standards, a minimum performance threshold is required for incentives:
+
+- **Minimum Score**: **0.6**
+- This threshold ensures that miners identify the main automation vectors (Selenium) and maintain high human safety while allowing for minor misses in other categories.
+
+---
+
+## Similarity & Time Decay
+
+- **Similarity check:** Submissions are compared against other SDKs; high similarity incurs penalties.
+- **Score decay:** Scores decay over 15 days to incentivize refreshed heuristics.
+
+### Example
+
+Assume:
+
+- 8 frameworks total
+- Perfect human accuracy → 1.0
+- Perfect protocol accuracy → 1.0
+- 6 frameworks detected perfectly → 6.0 points
+
+```math
+Final Score = (1.0 + 1.0 + 6.0) / 10
+            = 8.0 / 10
+            = 0.8
+```
+
+Any excessive human misclassification, protocol misses, or missing Selenium detection would reduce this to **0.0**.
+
+---
+
+## Submission Guide
+
+To build and submit your solution, please follow the [Building a Submission Commit](../../miner/workflow/3.build-and-publish.md) guide.
+
+## Submission Templates
+
+Templates and building instructions can be found in the [ADA Detection repository](https://github.com/RedTeamSubnet/ada-detection-challenge/tree/main/templates/commit).

src/redteam_core/challenge_pool/active_challenges.yaml

@@ -1,7 +1,7 @@
-ada_detection_v1:
-  name: "ada_detection_v1"
+ada_detection_v2:
+  name: "ada_detection_v2"
   description: "Detect anti-detect automation frameworks"
-  challenge_incentive_weight: 0.5
+  challenge_incentive_weight: 0.6
   challenge_image: "redteamsubnet61/ada_detection:v1.0.0-251215"
   target: ada_detection.controller.AADController
   challenge_manager: ada_detection.challenge_manager.AADChallengeManager
@@ -17,7 +17,7 @@ ada_detection_v1:
     max_self_comparison_score: 0.7
     min_acceptable_score: 0.6
   challenge_container_run_kwargs:
-    name: "ada_detection_v1"
+    name: "ada_detection_v2"
     platform: "linux/amd64"
     privileged: true
     environment:
@@ -37,7 +37,7 @@ ada_detection_v1:
     mem_limit: "12g"
     network: "redteam_local"
     environment:
-      CHALLENGE_NAME: "ada_detection_v1"
+      CHALLENGE_NAME: "ada_detection_v2"
   protocols:
     challenger: "http"
     challenger_ssl_verify: false
@@ -52,7 +52,7 @@ ada_detection_v1:
 ab_sniffer_v5:
   name: "ab_sniffer_v5"
   description: "Detect the driver type of the framework"
-  challenge_incentive_weight: 0.5
+  challenge_incentive_weight: 0.4
   challenge_image: "redteamsubnet61/ab_sniffer:v5.0.0-251202"
   target: ab_sniffer.controller.ABSController
   challenge_manager: ab_sniffer.challenge_manager.ABSChallengeManager

src/redteam_core/challenge_pool/controller.py

@@ -699,11 +699,6 @@ def _exclude_output_keys(self, miner_output: dict, reference_output: dict):
         """
         pass
 
-    @abstractmethod
-    def start_challenge(self):
-        """Initiates the challenge lifecycle by setting up and executing the challenge Docker container."""
-        pass
-
     @abstractmethod
     def _score_miner_with_new_inputs(
         self, miner_commit: MinerChallengeCommit, challenge_inputs

src/redteam_core/validator/models.py

@@ -114,6 +114,26 @@ def public_view(self) -> "MinerChallengeCommit":
             accepted=self.accepted,
         )
 
+    def remove_lower_than_highest_score(self):
+        """Remove all comparison logs that have scores lower than the highest score across all logs."""
+        if not self.comparison_logs:
+            return
+
+        highest_score = self.get_higest_comparison_score()
+
+        if highest_score == 0.0:
+            return
+
+        filtered_logs = {}
+        for ref_commit, logs in self.comparison_logs.items():
+            highest_scoring_logs = [
+                log for log in logs if log.similarity_score == highest_score
+            ]
+            if highest_scoring_logs:
+                filtered_logs[ref_commit] = highest_scoring_logs
+
+        self.comparison_logs = filtered_logs
+
     def remove_redundant_logs(self):
         """This is temporary function to clear scoring logs except in index 0 and comparison logs except the highest similarity score log."""
         if len(self.scoring_logs) > 1:
@@ -127,7 +147,7 @@ def remove_redundant_logs(self):
                 self.comparison_logs[ref_commit] = [highest_log]
 
     def get_higest_comparison_score(self) -> float:
-        """Get the minimum and maximum similarity score from all comparison logs."""
+        """Get the maximum similarity score from all comparison logs."""
         if not self.comparison_logs:
             return 0.0