Faithfulness

Evaluates whether claims in the output factually align with the contents of the provided context and input by checking for contradictions, ultimately ensuring factual consistency.

Input

• Required Inputs:
  • input: The user’s query or prompt
  • output: The model-generated answer to be evaluated
  • context: The source material that the output should be faithful to
• Optional Inputs
  • history: Previous conversation context

Output

• Result: Value in the continuous range [0, 1]
• Reasoning: Detailed explanation of faithfulness assessment

Interpretation

• Higher scores (closer to 1): Better faithfulness - most or all claims in the output are consistent with the provided context, input, and history
• Lower scores (closer to 0): Poor faithfulness - many claims in the output contradict or are not supported by the provided context, input, and history