When to Retrain after Drift: A Data-Only Test of Post-Drift Data Size Sufficiency
CALIPER is a data-only test designed to estimate how many post-drift samples are needed before retraining a model is stable. It works by checking if a sliding window of data exhibits sufficient state dependence using an effective sample size gate and a monotone locality test during weighted local regression. The method provides a principled way to decide when enough new data has accumulated for safe adaptation.