content-reliability-assessment-browser-extension

Content Reliability Assessment Browser Extension

📖 Overview

This project is a browser extension designed to help users systematically assess the reliability of online content. It automates a Multi-Layered Framework that evaluates content based on provenance, argumentation, and forensic integrity.

Current Status: The Proof of Concept (POC) is complete. The system currently performs robust text-only analysis, utilizing Large Language Models (LLMs) to classify content (Factual, Opinion, Fiction) and assess its logical coherence and rhetorical style.

🎯 Core Features (POC Complete)

The following features have been implemented and validated:

• LLM-Driven Analysis: A functioning analytical engine powered by Gemini (Cloud) or Mistral (Local/Ollama).
• Content Classifier: Successfully categorizes text into Factual, Opinion, or Fiction.
• Scoring Engine: Applies specific reliability criteria from our rubrics to generate a quantitative score.
• User Interface: A fully integrated browser extension popup that displays color-coded reliability ratings (Blue/Green/Purple) and detailed breakdowns.

⚙️ Tech Stack & Configuration

The extension supports a dual-mode backend for flexibility and privacy:

Mode	Engine	Requirement	Use Case
Cloud	Google Gemini	User API Key	High performance, requires internet connection.
Local	Mistral	Ollama	Privacy-focused, offline capability, runs locally.

🚀 Getting Started

Installation

1. Build the extension:

   npm install
   npm run build:chrome    # For Chrome/Edge
   npm run build:firefox   # For Firefox
   

2. Load in your browser:
   • Chrome/Edge: Load dist-chrome folder as an unpacked extension (Developer mode)
   • Firefox: Load dist-firefox folder as a temporary add-on
3. Configure your AI model:
   • Click the extension icon and open Settings
   • Choose between Cloud (Gemini) or Local (Mistral/Ollama)

Option 1: Cloud Mode with Gemini (Easiest)

1. Get a free API key from Google AI Studio
2. Enter your API key in the extension settings
3. Select “Gemini” as your model
4. Start analyzing pages

Option 2: Local Mode with Ollama (Privacy-Focused)

Run AI models completely locally without sending any data to external servers.

Quick Setup (Windows):

# Install Ollama
winget install Ollama.Ollama

# Pull the Mistral model
ollama pull mistral

# Configure CORS (PowerShell as Administrator)
[System.Environment]::SetEnvironmentVariable('OLLAMA_ORIGINS', '*', 'Machine')

Then restart Ollama and select “Mistral (Local - Ollama)” in extension settings.

📖 Full Setup Guide: See docs/OLLAMA_SETUP_WINDOWS.md for complete instructions, troubleshooting, and alternative models.

Benefits of local models:

• ✅ Complete privacy - data never leaves your computer
• ✅ No API key required
• ✅ Works offline after initial download
• ✅ No usage limits or costs

📚 Documentation & Frameworks

This tool is built upon rigorous information literacy frameworks. We highly recommend contributors read the core documentation located in the docs/ directory:

Document	Description
The Core Framework	The theoretical basis for the entire project, covering provenance and forensics.
Factual Rubric	Scoring criteria for evidentiary integrity and citation quality.
Opinion Rubric	Criteria for assessing logic, rhetorical style, and transparency.
Fiction Rubric	Guidelines for transparency in satire and creative writing.
Ollama Setup Guide	Complete guide for setting up local Mistral model on Windows.
Privacy Policy	How we handle your data, third-party services, and your privacy rights.

🗺️ Roadmap: Towards the Full Framework

With the text-only POC complete, development now shifts to the full implementation of the Multi-Layered Framework (specifically Part I and Part III). The goal is to evolve the system from a single analyzer into a Multi-Agent Orchestrator.

✅ Completed: Phase 1 (POC)

• Internal text analysis (Logic, Rhetoric, Cognitive Style).
• Content Classification (Factual/Opinion/Fiction).
• Basic UI and LLM integration.

🚧 Current Focus: Phase 2 - Provenance & Lateral Reading

Implementing Part I of the Framework: “Foundational Frameworks for Source and Provenance Evaluation”

• Lateral Reading Agent: Automate the “SIFT” method (Stop, Investigate, Find, Trace).
• Authority Verification: Integrate APIs (e.g., Google Scholar, Media Bias Fact Check) to verify author credentials and publisher reputation.
• Citation Analysis: Automated checking of links and references to detect “reference rot” or misrepresentation.

🔮 Future: Phase 3 - Multimedia Forensics

Implementing Part III of the Framework: “Advanced Multimedia and Synthetic Content Forensics”

• Metadata Analysis: Extraction of EXIF and other file headers.
• Synthetic Media Detection: Integration of deepfake detection models for images and video.
• C2PA Verification: Implementation of a scanner for Content Credentials and digital watermarks.

🧩 Final Stage: Phase 4 - The Orchestrator

Implementing Section 4.2: “Operationalizing the Framework”

• Decomposition Engine: Breaking down a URL/Page into text, image, and metadata components.
• Agent Synthesis: Merging outputs from Provenance, Argumentation (Text), and Forensic agents into a single, cohesive reliability report.

🤝 Contributing

We are looking for contributors to help us tackle the new roadmap phases:

1. Full-Stack Developers: To build the “Lateral Reading” agent that can perform live web searches and scrape results.
2. AI/ML Engineers: To assist with integrating computer vision models for Phase 3 (Multimedia Forensics).
3. Prompt Engineers: To refine the “Argumentation Agent” and ensure it works in harmony with the new agents.

How to Collaborate

1. Read the Multi-Layered Framework V3 to understand the full vision.
2. Fork the repository.
3. Create a feature branch (git checkout -b feature/LateralReading).
4. Commit your changes.
5. Open a Pull Request.

⚖️ License

Distributed under the GNU Affero General Public License v3.0 (AGPL v3). See LICENSE for more information.

```

This site is open source. Improve this page.