Deploying Machine Learning (ML) models in wireless networks comes with challenges beyond standard performance evaluation, particularly in their ability to remain reliable under changing network conditions. This study highlights some of these challenges by analyzing how models perform when tested on new radio links after training was completed, revealing significant performance degradation. A testbed was developed to study these effects, capturing real-world variations in wireless environments. Using this setup, the study based on link quality estimation (LQE), that is a critical aspect of network performance, demonstrates how data distribution shifts affect model performance. The findings from this first step emphasize the need for continuous monitoring and adaptation strategies, as well as further research on effectively implementing these methods in wireless networks.