The paper introduces the architecture-agnostic detection cost function (a-DCF), a metric designed to evaluate spoofing-robust automatic speaker verification (ASV) systems. The a-DCF generalizes the original detection cost function (DCF) used in ASV and is more flexible than the tandem detection cost function (t-DCF), which is restricted to cascaded speaker and spoof detectors. The a-DCF reflects the cost of decisions in a Bayes risk sense, with explicitly defined class priors and detection cost model. It can be applied to various classifier architectures, including those that produce a single output score. The paper demonstrates the effectiveness of the a-DCF through benchmarking evaluations of architecturally heterogeneous spoofing-robust ASV solutions, showing its ability to handle different evaluation metrics and architectures. The results highlight the advantages of the a-DCF over other metrics, such as the equal error rate (EER) and the t-DCF, in terms of flexibility, task-agnostic nature, and the ability to handle different evaluation setups.The paper introduces the architecture-agnostic detection cost function (a-DCF), a metric designed to evaluate spoofing-robust automatic speaker verification (ASV) systems. The a-DCF generalizes the original detection cost function (DCF) used in ASV and is more flexible than the tandem detection cost function (t-DCF), which is restricted to cascaded speaker and spoof detectors. The a-DCF reflects the cost of decisions in a Bayes risk sense, with explicitly defined class priors and detection cost model. It can be applied to various classifier architectures, including those that produce a single output score. The paper demonstrates the effectiveness of the a-DCF through benchmarking evaluations of architecturally heterogeneous spoofing-robust ASV solutions, showing its ability to handle different evaluation metrics and architectures. The results highlight the advantages of the a-DCF over other metrics, such as the equal error rate (EER) and the t-DCF, in terms of flexibility, task-agnostic nature, and the ability to handle different evaluation setups.