forecasting-ch16-prophetlisted

# Chapter 16: The Prophet ## Scope and intake Use when a business series has defensible calendar effects and changing trend suitable for Prophet. Do not use for treating an automatically fitted holiday component as a causal campaign effect. Ask only for unresolved material inputs: Which calendar events were known at the origin? Is their effect repeated in history? What trend flexibility is plausible? Which future regressors are actually supplied? ## Input contract and additional evidence A regular series with optional numeric regressor columns. Events go in config `events` as a list of `{name, date, lower_window, upper_window}` (future event dates are welcome). Regressors are named in config `regressors`; to forecast the future with regressors, append exactly `horizon` trailing rows with an empty `target` and the known regressor values. Minimal **format illustration**, not sufficient training data: ```csv timestamp,target,temp 2022-01-01,101.3,20.1 2022-01-02,103.7,20.9 2022-01-03,,21.4 ``` ## Executable interface Exact CLI columns: `timestamp,target[,regressors...]`. Supported method controls: `as_of, events, frequency, horizon, mode, origins, priors, regressors, season, seed, weekly, yearly`. Prophet with the event calendar as holidays and the declared regressors. `mode: auto` chooses multiplicative seasonality when a Box-Cox check on training data calls for a log scale, else additive. The changepoint prior is chosen from `priors` on `origins` earlier blocks; the