Run the Geneva Forecasting Engine on a time series and return forecast results.

Use this tool when the user has numerical time series data (e.g., monthly sales, daily temperatures, quarterly revenue) and wants to predict future values. The Geneva Expert System automatically selects the best forecasting model from 10 methods including exponential smoothing, Holt-Winters, and regression models.

Parameters

data : list[float] Time series observations. Minimum 3 data points, maximum 10,000. These should be sequential, evenly-spaced numerical values. horizon : int, optional Number of future periods to forecast. Default: one full seasonal cycle (e.g., 12 for monthly data). For monthly data, 18 is a good choice. For quarterly, 8. For weekly, 13. wave_periods : list[int], optional Seasonal cycle lengths. IMPORTANT — set this correctly for your data: - [12] for monthly data (default) - [4] for quarterly data - [52] for weekly data - [7] for daily data - [24] for hourly data - [1] for yearly/non-seasonal data confidence_level : float Prediction interval confidence (0.0 to 1.0). Default 0.95 gives 95% prediction intervals. Higher = wider bands, more confidence. method : int, optional Force a specific forecasting method (0–9). Omit to let the Expert System auto-select the best model. Methods: 0=LinearReg, 1-5=NonLinearReg, 6=SES, 7=DES (Double Exponential Smoothing), 8=HoltWinters, 9=Croston. Expert System (default) tries all and picks the best fit. seasonal_transform : int Seasonal transform to apply: 0=None (default), 1=Seasonal, 2=MPT (Moving Periodic Total). Use 1 or 2 for strongly seasonal data to improve forecast accuracy. smoothing : bool Enable median data smoothing (3-period window). Useful for noisy data. Default: False. max_periods_factor : float, optional Controls the fit window cap (nPPC × MPF). Higher values give the model more holdout data for evaluation, which can improve accuracy on long series. Default: engine default (1.5). Use 10+ for long series. Max: 100. holdout_ratio : float, optional Fraction of data reserved for model evaluation (e.g., 0.333). Default: engine default (1/3).

Returns

CallToolResult Contains a text summary with forecast values, accuracy metrics, model info, and prediction intervals.