Operational evidence standards for machine learning in wastewater treatment

Machine learning studies in water treatment often report high retrospective accuracy but rarely show whether models can support operational decisions. We analyse 423 full-text studies and a bounded six-plant wastewater-treatment scenario. Plant deployment (2.8%), real-time testing (5.2%), future-facing validation (18.2%) and uncertainty reporting (8.5%) remain uncommon. We propose an operation-aware evidence standard linking sensing context, validation timing, uncertainty and control role to deployment claims.