​
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