Forecasting use cases
Match models and evaluation to demand planning, energy forecasting, monitoring, and maintenance.
Demand and energy
Demand planning must distinguish observed sales from censored demand and account for promotions and hierarchy. Energy forecasting needs domain-specific inputs and historical vintages for load, renewables, prices, weather, and outages. Generic numerical accuracy does not replace the decision's physical and economic constraints.
Monitoring and maintenance
Forecast residuals, representations, dedicated anomaly models, and failure-risk models answer different questions. A widening uncertainty band is not automatically an anomaly or equipment-failure probability. Evaluate alert quality, lead time, false interventions, and operating regimes.
Choose a deployable evidence path
Use the browser playground for a target-only numerical baseline. Tasks requiring covariates, multiple targets, or different model heads need a suitable upstream implementation or external deployment. Preserve input and model provenance so downstream explanations can be checked.