Germany employs an iterative site selection process to narrow down candidate sites, culminating in a secure deep geological repository for high-level radioactive waste. This safety-oriented procedure requires dependable information and reliable, transparent decision-making. The decision-making process relies on physics-based numerical models, supported by lifecycle management ensuring sustainability and reproducibility of information products, e.g., uncertainty-informed hazard maps. This process requires a digital infrastructure capable of transforming data, i.e., geological and material properties, into predictive information for decision-making. Such an infrastructure must be modular and extensible to accommodate evolving data-acquisition protocols, support data validation, and ensure transparent, reproducible computational workflows. To address these research questions, we developed an actionable digital twin with a multi-layered architecture, where each layer provides the competencies necessary for a specific research contribution. The communication layer establishes a reproducible database foundation that provides uncertainty-informed data, intelligent data assembly capabilities, and extensible workflows for data updates. The model layer incorporates physics-based numerical models that yield, among others, spatiotemporal radionuclide concentration fields in the geological subsurface. Additionally, we employ surrogate models and Bayesian methods for uncertainty quantification and hazard assessment in a surrogate layer. The decision support layer, comprising benchmark scenarios, functions as an interface to provide actionable insights and feedback for subsequent decisions. The digital twin is demonstrated via a case study of radionuclide transport in a specific geological environment. It generates an uncertainty-informed hazard map, links data retrieval to scenario-based simulations, and integrates a surrogate model for computationally intensive tasks. Bayesian inference is employed to assimilate site-specific observations and analyze the impact of available measurements on the hazard assessment. At its core, the actionable digital twin hybridizes data and models for prediction, forensics, and introspection. It provides a transparent, reproducible, and modular framework that maps data into decision-relevant information, thereby supporting decision-making throughout the site selection process.