Artificial Intelligence Dose Prediction for Radiotherapy Planning Under Variable Linear Accelerator Commissioning Drift: A Conformal Calibration Evaluation of Healthcare Plan Quality

Artificial intelligence dose prediction models for radiotherapy planning are typically validated on commissioning data that remain fixed, yet linear accelerators drift in output and beam symmetry between calibration cycles. the present analysis formalises the resulting calibration problem as a distribution shift in the conditional dose distribution and proposes a conformal prediction wrapper that restores coverage guarantees under drift without retraining the underlying network. The wrapper is evaluated in a replication-based design that characterises the operating characteristics of the before-after-control-impact estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells 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 effect at 80% power, and that the minimum detectable effect at 80% power falls steeply as cluster count rises from 8 to 32. Conformal calibration is shown to be necessary because the intraclass correlation of 0.64 implies that cluster-level drift dominates residual variation, so naive coverage statements based on exchangeable calibration data are anti-conservative. The contribution is a formal coverage bound for split-conformal dose prediction under commissioning drift and a design table that links cluster count to detectable plan-quality degradation.