Code, processed data, and results for AutoML platform stress tests in environmental monitoring

This record provides the experiment scripts, the exact processed input CSVs, run configurations, result registries, retained test-set predictions, and figure code for the DAMDID/RCDL 2026 paper "Experimental Analysis of AutoML Platform Vulnerabilities in Environmental Monitoring Tasks and Risk Mitigation Strategies." Auto-sklearn 0.15.0, FLAML 2.5.0, and AutoGluon 1.5.0 were evaluated in single-seed regression runs on Beijing Multi-Site Air Quality (PM2.5), UCI Air Quality (CO), and a ten-station AURN subset (NO2) under Gaussian noise, MCAR missingness, reduced training samples, reduced search budgets, a combined noise-and-budget scenario, and a 2022–2023/2024 calendar split. RMSE and MAE in the result registries can be recomputed from the stored predictions; the included verification script checks this for every run. The README documents the software environments, rerun commands, source attributions, and known deviations of the processed inputs.