forecasting-ch03-weatherlisted

# Chapter 3: The Weather That Could Be Computed ## Scope and intake Use when seasonal structure or sensitivity to initial conditions needs to be understood before choosing a forecast. Do not use for operational weather prediction from the Lorenz teaching system. Ask only for unresolved material inputs: What interval is sampled? Which seasonal periods are physically or commercially plausible? Is the goal descriptive decomposition or future forecasting? Are current state estimates uncertain? ## Input contract and additional evidence CSV timestamp,target on a regular unique calendar; metadata frequency, season,units and cutoff. For a dynamics experiment separately specify the equations, initial states and integration tolerance; those are not columns in the business-series CSV. Minimal **format illustration**, not sufficient training data: ```csv timestamp,target 2024-01-01,100 2024-02-01,110 ``` ## Executable interface Exact CLI columns: `timestamp,target`. Supported method controls: `as_of, frequency, horizon, season, seed`. STL is descriptive. At least three seasonal cycles are required. The adapter refits an earlier vintage and reports the maximum historical trend revision; it does not issue a weather or business forecast. ## Applied procedure 1. Check regular spacing, outliers and whether at least two cycles exist for a proposed seasonal period. Plot the history before interpreting its spectrum. 2. For description, fit STL and verify observed equals trend+seasona