AI algorithms are increasingly used in hiring, yet their opacity raises concerns about fairness and legal compliance. This study presents a post hoc fairness audit of a proprietary candidate-job matching algorithm deployed in the Italian labor market. Using real-world datasets with precomputed match scores, we evaluate disparities across gender, age, and region of residence. Our analysis addresses three research questions: whether match scores reflect demographic disparities, how group fairness metrics compare to distributional measures in detecting bias, and whether representational bias alone signals unfairness. We combine standard group fairness metrics, such as Disparate Impact, with the Wasserstein distance to assess distributional fairness. While conventional metrics suggest minimal bias in gender and age, we uncover significant regional disparities disadvantaging Southern candidates with the help of distributional based metrics such as Wasserstein distance. These findings demonstrate the limitations of relying solely on group metrics and highlight the value of multi-metric approaches. Our methodology aligns with EU regulatory standards and offers a replicable framework for auditing opaque AI systems.