Why Busan planners see different seasonality than Seoul HQ models

National models assume interchangeable weeks. Coastal planners know typhoon season shifts dock labour and cross-town transfers in ways HQ templates miss. We start with geography-tagged history, then test whether disaggregated models outperform within statistically defensible sample sizes. When samples are thin, we blend regional and national estimates with explicit weights. Finance appreciates the transparency: every blended model publishes its shrinkage parameter and the review date. The work is tedious, yet it prevents the all-too-common pattern of blaming planners for “bad judgment” when the model never saw their weather reality.