forecast

Best practices for using the Chronulus MCP server to produce time-series forecasts with the NormalizedForecaster — demand, foot traffic, share of volume, seasonal weights, occupancy, or any quantity over a future horizon of hours, days or weeks, with no historical data required. Use whenever a user asks Claude to forecast, project, predict a trend over time, or plot a forecast with Chronulus in Claude.ai or Claude Code. Covers writing the session (situation + task), designing one reusable input_data_model, setting the horizon, reusing one agent across many items, rescaling the 0–1 forecast into real units with y_min/y_max (and invert_scale), showing the result with render_forecast_scorecard instead of hand-building a chart, and the optional on-demand risk scorecard.