This study addresses the challenge of selecting appropriate explainable artificial intelligence (XAI) methods by proposing a novel methodology to verify fidelity metrics using a transparent model, specifically a decision tree. The decision tree serves as a ground truth for explanations, allowing for perfect fidelity. The study aims to create the first objective benchmark for fidelity metrics, facilitating comparisons and surpassing existing methods. Two experiments were conducted using public datasets with 52,000 images, each with a size of 128 by 128 pixels. The results indicate that current fidelity metrics are unreliable, with the best metric showing a 30% deviation from perfect explanation. The study concludes that existing fidelity metrics are not suitable for real-world scenarios due to their sensitivity to out-of-distribution (OOD) samples. The proposed methodology can serve as a quality benchmark for future metric developments, emphasizing the need for new metrics that accurately capture the true fidelity of explanations, especially in fields like medical tasks.This study addresses the challenge of selecting appropriate explainable artificial intelligence (XAI) methods by proposing a novel methodology to verify fidelity metrics using a transparent model, specifically a decision tree. The decision tree serves as a ground truth for explanations, allowing for perfect fidelity. The study aims to create the first objective benchmark for fidelity metrics, facilitating comparisons and surpassing existing methods. Two experiments were conducted using public datasets with 52,000 images, each with a size of 128 by 128 pixels. The results indicate that current fidelity metrics are unreliable, with the best metric showing a 30% deviation from perfect explanation. The study concludes that existing fidelity metrics are not suitable for real-world scenarios due to their sensitivity to out-of-distribution (OOD) samples. The proposed methodology can serve as a quality benchmark for future metric developments, emphasizing the need for new metrics that accurately capture the true fidelity of explanations, especially in fields like medical tasks.