Artificial Intelligence Benchmarking of Antenatal Ultrasound Gestational Age Estimation Under Unmeasured Crown-Rump Length Operator Variability: A Healthcare Calibration-Transfer Evaluation

Antenatal ultrasound gestational age estimation depends on crown-rump length measurement, yet the operator variability that contaminates that measurement is rarely modelled when machine-learning estimators are benchmarked. the present analysis treats the problem as one of calibration transfer under an unmeasured covariate and characterises the operating characteristics of a before-after-control-impact design intended to detect estimator degradation when operator variability shifts between sites. No observational data are used or claimed. The analysis is a replication-based evaluation under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.64, with power estimated across 2000 design cells at cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design requires 52 clusters to detect a 5-percentage-point effect at 80% power. The intraclass correlation of 0.64 dominates the variance budget and makes cluster-level randomisation the binding constraint rather than within-cluster sample size. Benchmarking claims for gestational age estimators are not transferable across operator regimes unless the calibration-transfer design explicitly accounts for this clustering, and it derives the conditions under which a reported benchmark can be treated as invariant.