Reproducibility Data and Code for Energy-Aware TinyML Intrusion Detection on ESP32-S3: Model Selection under Latency, Footprint, and Board-Level Energy Constraints

This repository provides the supporting research data, model artifacts, source code, firmware, raw instrumentation records, and reproducibility documentation for the study “Energy-Aware TinyML Intrusion Detection on ESP32-S3: Model Selection under Latency, Footprint, and Board-Level Energy Constraints”, associated with manuscript IJIES 20266834. The deposit contains a single compressed archive comprising 309 files. Its contents include frozen TON_IoT-derived decision-tree implementations, model-depth variants, verified CICIoT2023 sample and checkpoint artifacts, C++ and ESP32-S3 firmware, external evaluation scripts, and original measurement records from an INA219-based board-level energy measurement system. The hardware evidence includes measurements of three single-tree classifiers and a five-tree random-forest comparator, together with serial records, GPIO2/CRC16 instrumentation, energy integration results, and independent audit reports. The V7 and V8 campaigns are documented as two separate powered sessions, with 16 accepted measurement slots and 48 synchronized workload phases. Earlier independently powered measurements are preserved separately. External evaluation materials cover IoT-23, IoT-SFC v2, and CTU-IDSEVAL-6. These results document cross-dataset sensitivity, differences in traffic composition, feature-availability constraints, and model behavior under distribution shift. They should not be interpreted as proof of a fully schema-equivalent, prospectively blinded external validation. The archive includes a file-level SHA-256 manifest, data-provenance documentation, evaluation scripts, model identity information, and reports describing the relevant experimental limitations. Board-level energy results refer to the measured power domain and do not establish independently traceable absolute current calibration or end-to-end intrusion-detection energy consumption. Original third-party datasets and attribution: CICIoT2023 is attributed to its original creators and the Canadian Institute for Cybersecurity. The IoT-SFC v2 and CTU-IDSEVAL-6 data retain their source attribution and applicable licensing conditions. The original TON_IoT dataset is referenced through its official source and recorded checksum rather than redistributed in this deposit. Protected device-memory backups and unqualified synthetic or aborted experiments are excluded from the public research archive. Archive filename: IoTD_IJIES_20266834_ZENODO_COMPLETE_V2_20261029.zip Archive SHA-256: f89d5144744aafffc9f3c6f78b6e1db710baff7c68354c201ed4b829a5a2a0a8 Related manuscript: IJIES 20266834. This deposit is intended to support independent inspection of the reported computations, model identities, experimental provenance, and reproducibility limits.