forecasting-ch12-benchmarkinglisted

# Chapter 12: The Competition ## Scope and intake Use when forecast methods need a fair chronological benchmark across series, origins and horizons. Do not use for claiming a global method ranking from one convenient synthetic panel. Ask only for unresolved material inputs: What horizon and loss match the decision? Which series and origins define deployment? Are weights business volume weights or equal-series weights? Is there an untouched final test? ## Input contract and additional evidence CSV timestamp,target and optional series_id; unique timestamp per series with regular frequency. Config horizon,season,origins and aggregation weights. Every compared method must face the same eligible targets. The rolling comparison needs 2 seasons + 4 horizons of history (72 monthly points for a 12-month horizon, 48 for six months, 36 for three); with less, the tool returns a provisional persistence baseline and says how many points are missing. Minimal **format illustration**, not sufficient training data: ```csv series_id,timestamp,target A,2025-01-01,100 A,2025-02-01,104 B,2025-01-01,20 ``` ## Executable interface Exact CLI columns: `timestamp,target`. Supported method controls: `as_of, conformal, country, criterion, frequency, future_regressors, horizon, origins, per_horizon_buckets, periods, pool, regressors, season, seed, transform`. `season` may be `"auto"` or omitted: the data profile then supplies it, and the profile (frequency, gaps, demand class, seasonal periods, o