Experimental Artefacts for HARDIO: A Health-Aware Reliable Detection-Imputation Orchestrator for Operational Data Quality in Small-Population WWTPs

Online sensor telemetry feeds every forecasting and control tool in wastewater treatment plants (WWTPs), but small-population plants below 2,000 population equivalent (PE) accumulate drift, flatlines, and outages that existing pipelines fail to catch because they evaluate detectors and imputers in isolation under inflated metrics. This work proposes HARDIO, Health-Aware Reliable Detection-Imputation Orchestrator, an end-to-end framework that combines a duration-weighted decay scorecard with a per-variable detector-imputer assignment driven by a four-pillar composite score: operational 35%, health coverage 30%, reconstruction 20%, and calibration 15%, together with a chained evaluation. HARDIO is validated as a deeply instrumented case study on the Cheles 1,177 PE WWTP, with 990,416 raw samples, 42 channels, and 432 days of monitoring. No uniform detector-imputer pair dominates: detection retains KNN Residual on 16 of 23 channels, SAITS on 5, LSTM-AE on 1, and RuleBased on 1, while chained imputation re-routes to LSTM-AE on 13, KNN on 9, and SAITS on 1. Chained R² exceeds 0.85 on 17 of the 19 channels with valid R², including inlet pH 0.952, outlet ammonium 0.923, outlet pH 0.935, and outlet dissolved oxygen 0.874. The four reactor- and clarifier-gas microsensors, CO₂, N₂O, NO, and NO₂, collapse the relative R² on near-constant signals and are therefore reported as 1−NRMSE ≥ 0.97. Against three uniform baselines, the orchestrated pipeline outperforms SAITS-only and KNN-Residual+KNN, with Wilcoxon p < 0.01 and bootstrap 95% CI strictly positive, and ties LSTM-AE-only in mean accuracy with a less catastrophic worst case. A 400-fold efficiency gap between RuleBased, 4.36 F1/s, and the deep models, 0.011 F1/s, drives an explicit edge-fog-cloud eligibility rule for future deployment. The released artefact exposes the model map, confidence labels, margins, and production masks.