How to Build a Synthetic Wastewater Demo Without Confusing It With Plant Validation

Why use synthetic data at all

Building an operator interface requires data that behaves like a plant: daily flow cycles, oxygen responses, storm events, sensor faults. Real plant data is often unavailable early, and when it is available it comes with confidentiality obligations.

Synthetic data fills that gap. It allows the full workflow — data review, quality checks, risk views, alerts, reports — to be designed, tested and demonstrated end to end.

Designing useful sample scenarios

The Waste Water Care MVP uses four deterministic scenarios: Normal Operation, Storm / High Flow, Aeration Degradation and Ammonia Risk. Each injects specific behaviours and specific data-quality artefacts, such as analyzer gaps during a storm or a dissolved-oxygen flatline during aeration degradation.

Determinism matters. The same scenario produces the same data every time, so reviewers see consistent results and tests can verify behaviour precisely.

What a synthetic demo can show

A synthetic demo can show that the workflow is coherent, that the interface communicates confidence and limitations clearly, that data-quality rules behave as specified, and that alerts and reports produce an auditable record.

What it cannot show

It cannot show that a forecasting model works on a real plant. Synthetic signals encode the assumptions of whoever generated them; a model that performs well on them has learned those assumptions, not the plant.

For that reason every predictive element in the Waste Water Care demo is labelled as curated demo output, sample data is labelled as such, and driver views are labelled illustrative.

Labelling as a design discipline

The risk with polished demos is that they are mistaken for evidence. Clear, persistent labelling — not hidden in tooltips — keeps the boundary visible. A model card that states "validated model: none" is more useful to a technical reviewer than an impressive chart without context.

Takeaway

Synthetic data is a tool for building and explaining a product. Plant validation is a separate, later step that requires real data, a benchmark and an agreed evaluation protocol.