Masked-Reference Field Validation Artifacts for a Training-Free Micro-Edge Ag-IoT Data-Quality Framework

This record contains the reproducibility artefacts for the manuscript “Masked-Reference Field Validation of a Training-Free Micro-Edge Framework for Agricultural IoT Data Quality.” The study evaluates a frozen, training-free, and resource-aware micro-edge framework for Agricultural Internet of Things (Ag-IoT) data quality using a masked-reference field validation protocol. Trusted observed field values were temporarily hidden, repaired using nine methods, and then compared with their stored references to support direct field-based repair scoring. The repository includes MATLAB scripts, mask manifest, fixed seeds, configuration files, result workbook, figure source data, composite figures, supplementary figures, literature-screening file, dataset-source notes, run-level logs, and reproducibility notes. These artefacts support the reported analyses on reconstruction accuracy, plausibility violations, confidence, fallback behaviour, and replay-based resource indicators. Raw third-party datasets are not redistributed in this record. The Field Validation Pool (FVP) records were derived from three raw agricultural observation datasets archived in Zenodo: Raw Tomato Greenhouse Environmental and Plant Growth Dataset (https://doi.org/10.5281/zenodo.20096294), Raw Rock Melon Environmental and Plant Growth Dataset (https://doi.org/10.5281/zenodo.20096467), and Raw Chinese Kale Environmental and Plant Growth Dataset (https://doi.org/10.5281/zenodo.20096516). Users who wish to rerun the full workflow should obtain the raw datasets from the original Zenodo records or data owners. The masked-reference procedure scores hidden observed field values only. It does not claim to know the true values inside natural missing field outages. Therefore, the artefacts should be interpreted as supporting materials for field-based validation, reproducibility, and transparency.