Developed ambiguity handling for the faithfulness metric evaluation in the confident-ai/deepeval repository, focusing on improving the reliability of contradiction reporting. The solution introduced logic to flag ambiguous verdicts when penalize_ambiguous_claims is enabled, ensuring that ambiguous claims are surfaced during evaluation. This approach enhanced the accuracy of model assessment by making the evaluation pipeline more robust to uncertain cases. The work was implemented using Python and leveraged data analysis and machine learning skills to refine metric calculations. By addressing ambiguity in verdicts, the update contributed to safer model deployment and provided more dependable metrics for evaluating model faithfulness.