TrustGate-IoT: Reproducibility Package for Purpose-Aware Governance of IoT-Derived Open Data under Changing Conditions

Abstract

This record provides the reproducibility package for TrustGate-IoT, an observation-level governance framework for IoT-derived Open Data under changing conditions. The package supports auditability and reproduction of two frozen evaluation studies. Study I provides controlled internal characterization using the UCI Appliances Energy Prediction dataset with separated Calibration, Validation, and Test stages, frozen corruption scenarios, clean contextual challenges, B0–B5 architectural comparators, paired multi-seed analysis, mechanistic ablation, hard-rule auditing, and runtime evaluation. Study II provides a zero-retuning external replication using 36 physical SensEURCity sensor nodes from Antwerp, Oslo, and Zagreb, with inference conducted at the physical-node level. The deposited archive contains the scientific implementation, frozen configuration and protocol artifacts, benchmark-construction and integrity manifests, official evaluation outputs, statistical evidence, figures, manuscript materials, licensing information, and append-only provenance records required to trace reported results to frozen inputs. Third-party raw datasets are not redistributed. The exact upstream datasets and cryptographic identities used in the study are documented in the included README_REPRODUCIBILITY.md: - UCI Appliances Energy Prediction: DOI 10.24432/C5VC8G- SensEURCity Version 3.0.0: DOI 10.5281/zenodo.17858205 The external replication is explicitly a zero-retuning evaluation. The primary comparison reveals a containment–retention trade-off rather than universal superiority: static purpose-aware governance provides stronger corrupted-data containment, whereas TrustGate-IoT provides higher clean-data retention in exploratory and analytical modes. The deposited archive has been independently verified by cryptographic hashes and contains no redistributed raw third-party data.