Evaluation of Intel’s Trust Domain Extensions (TDX) for Replicating and Validating Research in Management Science

Reproducibility of scientific research is vital for maintaining the integrity of academic disciplines. Currently, journals and conference organizers rely on varying standards and platforms to evaluate and ensure the reproducibility of research. Additionally, current methods using unprotected environments risk security and integrity issues. This project assesses Intel’s Trust Domain Extensions (TDX) framework, a CPU-level technology for implementing a trusted execution environment, as a potential standardized solution for replicating and validating research findings in the business, economics, and management fields. Specifically, we perform a comprehensive evaluation of replication packages from articles published in Management Science in 2024 using TDX, leveraging major providers such as Google Cloud and Microsoft Azure. Our analysis focuses on four core aspects: 1) performance, measured in terms of replication success rates and accuracy compared to the paper’s results 2) costs, including computational resource usage as a result of runtime; 3) identified best practices in replication methods and software environments; and 4) technical challenges encountered during replication, including data availability and software dependencies. Our team’s results yielded high success rates, with few failures stemming from unavailable datasets and software license restrictions. We also saw relatively low credit usage per paper replicated ($1.33), and, on average, around a 2-hour runtime, proving that TDX may be a dependable solution for replication purposes.