Faithfulness Scorer Example

This example demonstrates how to use Mastra's Faithfulness Scorer to evaluate how accurately responses adhere to the provided context without introducing unsupported information.

Prerequisites

• Node.js 22.13.0 or later
• pnpm (recommended) or npm
• OpenAI API key

Getting Started

1. Clone the repository and navigate to the project directory:

   bashCopy

   git clone https://github.com/mastra-ai/mastra
   cd examples/basics/scorers/faithfulness
   

2. Copy the environment variables file and add your OpenAI API key:

   bashCopy

   cp .env.example .env
   

   Then edit .env and add your OpenAI API key:

   envCopy

   OPENAI_API_KEY=sk-your-api-key-here
   

3. Install dependencies:

   bashCopy

   pnpm install --ignore-workspace
   

4. Run the example:

   bashCopy

   pnpm start
   

Overview

The Faithfulness Scorer evaluates how accurately responses adhere to provided context without introducing unsupported claims. It analyzes:

• Factual accuracy of claims against provided context
• Detection of unsupported or fabricated information (hallucinations)
• Context adherence and reliability
• Verification of all response claims against source material

Example Structure

The example includes three scenarios:

1. High Faithfulness: Testing Tesla Model 3 facts where all response claims are supported by context
2. Mixed Faithfulness: Testing Python programming language where some claims are supported and others are not
3. Low Faithfulness: Testing Mars information where response claims contradict the provided context

Each scenario demonstrates:

• Setting up the scorer with specific context and model configuration
• Providing context arrays for fact verification
• Running faithfulness evaluation on responses
• Interpreting the results with detailed reasoning about claim support

Expected Output

The example will output:

• The provided context, input query, and response for each scenario
• The scorer result with:
  • Score (0-1, where 1 indicates perfect faithfulness to context)
  • Detailed reasoning explaining which claims are supported or unsupported

Key Components

• createFaithfulnessScorer: Function that creates the faithfulness scorer instance
• Scorer configuration:
  • model: The language model to use for evaluation (e.g., OpenAI GPT-4)
  • options.context: Array of context strings that serve as the source of truth
  • options.scale: Optional scale factor for the final score (default: 1)
• scorer.run(): Method to evaluate input/output pairs for faithfulness
  • Takes { input, output } where:
    • input: Array of chat messages (e.g., [{ role: 'user', content: 'question' }])
    • output: Response object (e.g., { role: 'assistant', text: 'response' })
  • Returns { score, reason } with numerical score and detailed explanation