DeFedOblivio: Replication and Traceability Package

Replication and traceability package for the paper “DeFedOblivio: History-Based Committee-Governed Decentralized Federated Unlearning”. This archive contains the experimental artifacts used to support the evaluation reported in the paper, including experiment configurations, replay schedules, model architecture excerpts, dataset-partitioning code, assembled result files, plotting scripts, result-assembly utilities, membership-inference attack evaluation code, and pilot convergence analysis. The package is intended to support result traceability, figure regeneration, configuration-level reproduction, and independent reimplementation of the method described in the paper. It is not a full public release of the complete execution core. The underlying decentralized federated learning testbed includes proprietary/licensed components that cannot be redistributed. Nevertheless, the package provides the artifacts required to inspect how the reported results were configured, assembled, analyzed, and plotted. Included experiments cover full-depth unlearning and retained utility, membership-inference attack evaluation, rollback fidelity, rollback-depth sensitivity, and recalibration-availability stress tests across MNIST, CIFAR-10, and FEMNIST settings. Portions of the auxiliary code, including some formatting, plotting, result-assembly, and utility code, were drafted, enhanced, or reviewed with assistance from AI-based tools. The authors reviewed and validated the final artifacts and remain responsible for the package contents and interpretation of the results.