Interpretability Audits of Continuous Glucose Monitor Forecasting Models Under Sensor Warm-Up Calibration Gaps: A Healthcare Safety Evaluation

Continuous glucose monitor (CGM) forecasting models are typically trained and evaluated on sensor readings that have already passed the manufacturer's warm-up calibration window, yet deployed models must issue predictions during warm-up when the signal is systematically biased. The audit is specified as a before-after-control-impact (BACIP) estimator with a declared variance structure: between-cluster standard deviation 6.0, residual standard deviation 4.5, and intraclass correlation 0.64. Power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points.54 clusters are required to detect a 5 percentage point interpretability deficit at 80% power, and derives the minimum detectable effect curve for the BACIP interaction. The contribution is a reproducible audit protocol that separates warm-up bias from model-attributable error, together with the design parameters a healthcare safety evaluation would need to declare before deployment.