Forecast with ranges and scorecard — Atom

Enables AI clients to produce probabilistic forecasts with prediction intervals for numeric time series such as sales, demand, traffic, and sensor readings, using a panel of zero-shot foundation models or an accuracy-weighted ensemble. Supports up to 64 series per call, optional covariates, horizons up to 512 steps, and usage or credit checks, all without training or feature engineering.

Enables statistical forecasting and anomaly detection over numerical time-series data using Holt linear smoothing, multi-step projections, and variance-based confidence bands.

Enables demand forecasting and replenishment recommendations using statistical models (Syntetos-Boylan classification, AutoETS, TSB) and provides tools for forecasting, evaluation, and order quantity calculation.

Enables agents to pull live overnight forecasts for grower blocks and score frost risk with an itemised, versioned engine, returning cold-window start/end hours and a run-protection, stand-by, or no-action verdict. It also lets agents list blocks and sheets, create or update idempotent decision records, and replay their SHA-384 hash chain to verify nothing was altered.